Category: AI Agent Hosting

Hosting for the AI-agent era: where to run sites built with Claude, VPS setups for AI agents, WordPress MCP and AI-driven site operations.

  • Kinsta API + Claude: Automate Your WordPress Site Ops

    Kinsta API + Claude: Automate Your WordPress Site Ops

    Kinsta’s REST API, in public beta and available to all Kinsta customers as of October 2026, can run WP-CLI commands, update plugins, push staging to live, take manual backups, clear caches and report usage, which makes it a good target for scripts Claude writes. This guide builds three automations on it: bulk plugin updates across every site you run, a backup-push-and-clear-cache deploy step, and a weekly usage report.

    The endpoints below come from Kinsta’s API documentation and its API reference (OpenAPI specification version 1.111.0), checked in October 2026; this guide does not include our own tests.

    Kinsta’s reference is the authority here, and endpoints can change while the API is in beta. Each automation below names the endpoint it uses, so you can paste the matching reference page into Claude and have it confirm the current shape before you run anything.

    Why the Kinsta API is interesting

    The Kinsta API covers WordPress sites and environments, backups, site and edge caches, the CDN, analytics, domains and DNS. It can also run WP-CLI commands over HTTP, through POST /sites/environments/{env_id}/run-wp-cli-command. That last piece means you’re not limited to the buttons in MyKinsta: WP-CLI, which WordPress.org says can do “anything you can do in the WordPress admin,” becomes something a script can run on any of your sites. One limit to know: the API only accepts commands that start with wp and use letters, numbers, spaces and a few punctuation marks, so no pipes or chained shell commands.

    Pair that with Claude writing the scripts and you get what this site calls AI agent hosting. If that phrase is new to you, the primer on what AI agent hosting actually means is the place to start. The short version: the host provides clean, documented control surfaces; the agent does the tedious work; you review and approve.

    For context, Kinsta plans start at $35 a month for one site, Kinsta routes traffic through Cloudflare with Enterprise DDoS protection, and it now operates on Oracle Cloud Infrastructure, which replaced its previous Google Cloud setup. Kinsta says its API is available to all customers while it remains in public beta, which is not true of every managed host.

    What you need before you start

    • A Kinsta API key. In MyKinsta, go to your username > Company settings > API Keys, click Create API Key, choose an expiration, name it and generate it. Kinsta shows the key only once. Company owners, administrators and developers can create keys, and a key’s access depends on the user who creates it. Keep it in an environment variable, never in a script you commit.
    • Your company ID. The site-listing call needs it as a parameter. It’s the idCompany value in the MyKinsta URL when you open a site, and the environment ID appears in the same URL after the site ID.
    • A terminal with curl and jq. That’s the whole toolchain. The API takes and returns JSON, so no SDK is required.
    • Claude. Claude Code can run and debug the scripts itself; the chat interface works fine if you’d rather paste code around.
    • A staging environment to practice on. Non-negotiable. Give every automation below its first run against staging, not a live client site.

    One workflow note: everything below matches Kinsta’s specification as of October 2026, but the API is in beta. Paste the relevant reference page into Claude and let it confirm the current shape before you run anything.

    Automation 1: bulk plugin updates across every site

    This is the one that pays for the setup time. The WP-CLI endpoint takes an environment ID in the path and a wp_command in the body, and returns an operation_id because execution is asynchronous. The core loop looks like this:

    # set KINSTA_KEY in your shell first; never commit it
    API="https://api.kinsta.com/v2"
    
    # every environment you want to patch
    ENVS="ENV_ID_ONE ENV_ID_TWO ENV_ID_THREE"
    
    for ENV in $ENVS; do
      curl -s -X POST "$API/sites/environments/$ENV/run-wp-cli-command" -H "Authorization: Bearer $KINSTA_KEY" -H "Content-Type: application/json" -d '{"wp_command": "wp plugin update --all"}' | jq -r '.operation_id'
    done

    A prompt along these lines gets Claude to build the rest: “Write a bash script that runs wp plugin update –all on these Kinsta environment IDs via the run-wp-cli-command endpoint, polls each operation until it finishes, and prints a pass/fail table.” Ask for rate-limit handling up front. Kinsta allows 120 requests per minute per company and 5 per minute for resource creation, reports your status in RateLimit response headers, and says the limit can’t currently be raised. When you poll GET /operations/{operation_id}, a 202 means still running, a 200 means finished, and a 500 means the operation failed. Expect to review and iterate on the first draft.

    For an urgent security fix, swap the command for a single plugin, wp plugin update the-vulnerable-one, and the same loop patches every site in the list. Kinsta also has a dedicated endpoint, PUT /sites/environments/{env_id}/plugins/bulk-update, that takes a list of plugin names if you’d rather not go through WP-CLI.

    Automation 2: backup, staging push and cache clear on demand

    The usual manual routine is: update on staging, check it, push to live in MyKinsta, then remember to clear the cache. The API removes the remembering. Clearing the site cache is a single call; note that the environment ID goes in the request body, not the path:

    curl -s -X POST "$API/sites/tools/clear-cache" -H "Authorization: Bearer $KINSTA_KEY" -H "Content-Type: application/json" -d '{"environment_id": "'"$ENV"'"}'

    A sensible deploy script chains four documented steps. Take a manual backup of the live environment with POST /sites/environments/{env_id}/manual-backups (an optional tag labels it). Push staging to live with PUT /sites/{site_id}/environments, passing source_env_id and target_env_id; by default the push copies the database and files and runs search and replace. Poll /operations until each step returns 200. Then clear the live cache. The backup step matters: a push is the one action in this post that can overwrite your live site, and the manual backup gives you a restore point through POST /sites/environments/{target_env_id}/backups/restore. Put a confirmation prompt in the middle; automation doesn’t mean removing the human, it means removing the busywork around the human.

    Automation 3: the Monday usage report

    Schedule a weekly cron job that pulls the site list for your company ID (GET /sites?company=...), then this month’s visits and server bandwidth for each site (GET /sites/{site_id}/usage/visits/this-month and /usage/bandwidth/this-month), writes a Markdown table, and hands it to Claude to summarize into three bullet points. jq does the formatting. Per-environment analytics endpoints add top client IPs and a response-code breakdown if you want more detail.

    Boring? Completely. But Kinsta plans are billed on either visits or server bandwidth, and on visit-based plans Kinsta charges $0.50 per 1,000 visits over the cap. A weekly report shows a site drifting toward its allowance before the overage lands on an invoice, and the top client IPs endpoint helps you spot a single source hammering a site. Kinsta’s own plan limit alerts email you at 80% and 100% of usage; the report adds the trend.

    Where Claude fits in: two modes

    Everything above is mode one: Claude writes scripts, you run them on a schedule or on demand. It’s the right default because the output is inspectable: you can read a bash script before it touches production.

    Mode two is letting Claude drive your sites conversationally through an MCP server, where you type “update the plugins on the Henderson site and clear the cache” and it happens. That’s a bigger topic with its own trade-offs, covered separately in managing WordPress with Claude and MCP. If MCP is an unfamiliar acronym, start with the plain-English explainer. A sensible rule: scripts for anything destructive, conversation for anything read-only, and no unsupervised environment pushes.

    Downsides

    • Everything is asynchronous. Long-running actions return an operation ID you have to poll. Claude handles the boilerplate, but your scripts are longer and slower than you’d expect.
    • Key access follows the person who creates it. Kinsta says a key’s access depends on the creating user’s role, so a key made by an owner or administrator can do what they can, including deleting sites. Have a company developer create keys where that’s enough, set an expiration, and revoke keys when someone leaves the company.
    • Rate limits are real. 120 requests a minute per company is plenty for a short loop, and worth engineering around if you manage hundreds of sites.
    • Premium pricing, no email. At $35 a month for one site ($350 a year on annual billing), Kinsta only makes sense if the automation saves real billable hours, and Kinsta says “we don’t provide email as part of our hosting plans,” so budget for email separately.

    If the price is the blocker, Cloudways starts at $24 a month for the 2GB server it recommends for live sites, its Copilot assistant monitors server health and explains problems in plain terms, and Cloudways documents its own MCP server. Cloudways vs Kinsta compares the two directly, and the Copilot guide shows what Copilot does day to day. For the wider field, the best AI web hosting roundup compares hosts on how ready they are for agent workflows.

    Frequently asked questions

    Can the Kinsta API run WP-CLI commands?

    Yes. POST /sites/environments/{env_id}/run-wp-cli-command takes a wp_command (for example wp core version), starts it asynchronously, and returns an operation ID you poll at /operations. Commands must start with wp and use a limited set of characters, so shell pipes and chained commands aren’t accepted.

    Is the Kinsta API free to use?

    The API costs nothing extra. Kinsta lists it under “Premium features included in all plans,” which start at $35 a month for one site, or $350 a year billed annually, as of October 2026. It’s in public beta and available to all Kinsta customers. You’re paying for the hosting; the automation surface comes with it.

    Do I need to know how to code to automate Kinsta with Claude?

    You need to be able to read a short bash script and run commands in a terminal; that’s the realistic floor. Claude writes the code, but you should understand roughly what a script does before pointing it at production. If you can follow the examples in this post, you’re ready.

    Is it safe to run these automations on production sites?

    With guardrails, yes. Run every automation on staging first, take a manual backup before anything destructive, and put a human confirmation step inside any script that pushes environments. Read-only automations like the usage report carry little risk beyond keeping the key safe.

    What’s the difference between the Kinsta API and an MCP server?

    The API is the raw HTTP surface: you or your scripts call it directly. An MCP server is a translation layer that lets an AI assistant call capabilities conversationally. Scripts are more auditable; MCP is more convenient. You can use both: scripts for changes, conversation for questions.

    Last checked: October 2026. Prices and plans change; check the vendor’s page before you buy.

    Want the guided version? Our free video courses are in production, starting with “Launch Your First Website with Claude.”

  • Where to Host a Website Built With Claude Code

    Where to Host a Website Built With Claude Code

    Where to host a website built with Claude Code depends on what it built. As of October 2026, a static site can live free on Netlify’s Free plan or Vercel’s Hobby plan (non-commercial use only), or on Hostinger from $2.99 a month; a Node.js app needs a host that runs Node, such as Hostinger from $3.99 a month or Cloudways Velocity from $20; and a site non-developers will edit belongs on WordPress.

    This guide is based on each host’s published plans and documentation, checked in October 2026; it does not include our own tests.

    Claude Code will build you a website, write the CSS and fix its own bugs. What it won’t do is tell you where that site should live. You end the session with a folder of files and a hosting decision. This is the decision tree: static site, dynamic app, or WordPress conversion, with current prices and the cases where a free tier is all you need. If you haven’t started the build yet, read can Claude build a website first; this guide picks up where that one ends.

    First, figure out what Claude actually built

    Start here, because the answer decides everything downstream. Open the project folder and look at what’s actually in it.

    • Static site. You see index.html, a css folder, maybe some JavaScript, and no server code. Portfolios, landing pages and small business sites often end up like this. It’s also the cheapest kind of site to host.
    • Dynamic app. There’s a package.json with a start script, an Express or Next.js server, an app.py, or a database file. Something has to stay running to serve requests. You need a host that runs your language.
    • CMS-shaped project. The files are static today, but the plan involves a blog, weekly content changes, or a business partner who will never open a code editor. That’s a content management problem, and the usual answer is WordPress.
    What it looks likeWhere to host itMonthly cost (October 2026)
    Static siteHTML/CSS/JS only, no server codeNetlify Free, Vercel Hobby (non-commercial), Sevalla, Hostinger$0 on the free tiers; Hostinger from $2.99 on a 48-month prepay
    Dynamic appNode.js or Python backend, databaseNode.js: Hostinger or Cloudways Velocity. Python: a VPS or an app platformHostinger from $3.99 on a 48-month prepay; Velocity from $20
    Needs a CMSNon-coders will edit it regularlyWordPress on Hostinger or KinstaHostinger from $2.99; Kinsta from $35

    Hosting a static Claude Code site

    Quick verdict: if Claude built you plain HTML and CSS, you should be paying somewhere between nothing and the price of a coffee.

    The free tiers on Netlify and Vercel can host it, with limits worth reading. Netlify’s Free plan is $0 with a 300-credit monthly limit; bandwidth costs 20 credits per GB and each production deploy 15 credits, so the plan covers at most 15 GB of bandwidth in a month, less once you count deploys. When you hit the limit, Netlify pauses your projects until the next billing cycle. Vercel’s Hobby plan includes 100 GB of Fast Data Transfer a month, but Vercel restricts it to “non-commercial personal use only,” so a business site belongs on a paid plan. Both deploy from a Git repository, and Netlify also accepts a project folder dragged into its Deploy zone. The trade-offs: no email, and your site lives on infrastructure whose pricing and policies you don’t control.

    Kinsta’s static site hosting has moved. Kinsta removed application, database and static site hosting from MyKinsta on February 2, 2026, and those services now run on Sevalla, which Sevalla’s site calls “a Kinsta product.” Sevalla’s static hosting is $0 for up to 100 sites, with 600 build minutes and 100GB of bandwidth a month per account, deployed from GitHub, GitLab or Bitbucket. It’s managed from the Sevalla dashboard, separate from Kinsta’s WordPress hosting. And if you want the everything-included route, Hostinger‘s Premium plan costs $2.99 a month on a 48-month prepay ($143.52 up front), renews at $10.99, and bundles the thing the free tiers never give you: a domain free for the first year and 2 mailboxes per website, also free for the first year. Upload the folder with Hostinger’s File Manager, or use its Git integration, which covers “custom PHP and HTML websites” on web and cloud plans.

    Hosting a dynamic app

    If Claude built you a Next.js app with API routes, an Express backend, or a Python service with a database, the free-tier math changes. Free plans meter usage and stop when you hit the limit: Netlify pauses every project on the account, and on Vercel Hobby you generally wait until 30 days have passed. That’s a poor fit for an app people depend on.

    For a Node.js app, two managed options are documented. Hostinger runs Node.js web apps on its Business (now sold as Unlimited) and Cloud plans, advertised “From $3.99/mo, no usage fees”; Unlimited is $3.99 a month on a 48-month prepay and renews at $16.99. It supports frameworks including Next.js, Express.js, NestJS and Fastify, deploys from GitHub with automatic builds on every push or from a ZIP, and its Hostinger Connector lets Claude Code deploy for you. Cloudways sells managed Node.js hosting as Cloudways Velocity: Starter is $20 a month for 2GB RAM, 2 vCPUs and 5 apps, deploys from Git, can provision PostgreSQL, and includes OS patching and SSL. Cloudways’ main Flexible platform is built for “WordPress, Magento, Laravel & PHP apps,” so a PHP or Laravel backend fits there, from $24 a month for the 2GB server Cloudways recommends for live sites, and Copilot, Cloudways’ AI assistant, helps diagnose server problems.

    Python is the gap: neither Hostinger’s web plans nor Cloudways documents Python app hosting. For a Python backend, use an app platform such as Sevalla, which publishes Python quick starts for Django and FastAPI, or run your own VPS. We’ve written up how to set up a VPS for your AI agents if you want that control, but be honest about whether you’ll actually apply security updates every month. That’s the whole argument for managed.

    If you’re weighing the budget end against the managed end, our Hostinger vs Cloudways comparison covers that fork in detail.

    When WordPress is the smarter move

    A common scenario: someone ships a clean static Claude build, and a few weeks later their business partner asks how to change the pricing table. The answer, “open a code editor, edit the HTML, redeploy,” doesn’t work for someone who doesn’t code.

    If non-developers will edit the site, or you’re planning a blog that publishes more than once a quarter, convert the build to WordPress. Claude Code can do much of this: point it at the static files and ask it to turn them into a WordPress theme. Then the hosting choice is a budget question. Hostinger gets a converted site live for a few dollars a month. For a site that earns money, managed WordPress such as Kinsta, from $35 a month for one site, adds staging, daily backups and a REST API that Kinsta says is available to all customers while in beta. One caveat: Kinsta doesn’t host email, so keep your mailboxes at your registrar or on Google Workspace. If budget is the whole decision, our cheapest AI website builder roundup covers the low end.

    If you’d rather keep editing with Claude after the conversion, InstaWP offers WordPress hosting with an MCP server built in (it calls the feature InstaMCP), so Claude or Cursor can work on the live site through chat. Sites start as $2-a-month sandboxes and move to managed hosting from $5 a month per site, per its pricing page.

    The deployment workflow that actually works

    Whatever host you pick, put the project in a Git repository first. Not for engineering purity: because it turns every future update into “push and done” instead of “which files did I change?”

    • Initialize the repo. Ask Claude Code to run git init, write a sensible .gitignore, and push to GitHub. Anthropic says Claude Code “stages changes, writes commit messages, creates branches, and opens pull requests.”
    • Connect the host. Netlify, Vercel, Sevalla and Hostinger’s Node.js web apps all deploy from the GitHub repo on every push. For PHP and HTML sites, Hostinger’s Git integration and Cloudways’ Git deployment pull from your repository, or copy files over SSH, which Cloudways provides.
    • Let Claude do the deploying. This is the part people miss: Claude Code runs commands. It can execute the deploy itself, verify the live site responds, and fix what broke. Ask it to write a deploy script and a README so future-you knows how the whole thing works.

    That last point is why hosts with APIs and agent connectors matter: it’s the same property that matters for agent workflows generally, which the best AI web hosting guide compares.

    Common mistakes to avoid

    • Buying managed WordPress hosting for a static brochure site. That’s paying $35 a month, the entry price at both Kinsta and WP Engine, for a job a free tier does. Match the hosting to what Claude built, not to what the ads say.
    • Leaving the site on a free subdomain. yourbusiness.vercel.app looks temporary, and Vercel’s Hobby plan is for non-commercial use anyway. Buy a domain on day one; Hostinger’s Premium plan on a 48-month term includes one free for the first year.
    • Pushing API keys to a public repo. Generated code can put keys into client-side JavaScript. Search the project for anything that looks like a secret before your first push.
    • Running a DIY VPS with no backups. If you go the raw-server route to save money, the first thing you configure is automated backups. Managed platforms do this for you; that’s largely what you’re paying for.

    Frequently asked questions

    Can I host a website built with Claude Code for free?

    If it’s static (HTML, CSS, JavaScript, no backend), yes. Netlify’s Free plan, Vercel’s Hobby plan (non-commercial use only) and Sevalla’s static hosting all cost $0 within their limits. Budget for a custom domain regardless. Apps with a backend are where free stops making sense; plan on a paid host that runs your language, such as Hostinger from $3.99 a month or Cloudways Velocity from $20 for Node.js.

    Do I need special hosting for an AI-generated website?

    No. A website Claude built is made of the same files as a website a human built, and any normal host serves it. The question that matters is whether it’s static files or a running application, and for an app, which language it runs. That decides the hosting type and the price.

    Can Claude Code deploy the website by itself?

    Yes. Claude Code runs command-line tools, so it can push to GitHub, trigger a deploy, rsync files to a server over SSH, and then check that the live URL responds. Scope any credentials you give it to the one project. It’s one of the main reasons to use Claude Code over a chat-only AI for building sites.

    Should I convert my Claude Code site to WordPress?

    Only if people who don’t code will edit it, or you’re publishing content on a schedule. For a set-and-forget landing page, WordPress adds maintenance you don’t need. For anything with a blog or a non-technical editor, the conversion pays for itself the first time someone updates a page without touching the code.

    Last checked: October 2026. Prices and plans change; check the vendor’s page before you buy.

    Want the guided version? Our free video courses are in production, starting with “Launch Your First Website with Claude.”

  • Manage WordPress With Claude: WordPress MCP Tutorial

    Manage WordPress With Claude: WordPress MCP Tutorial

    To manage WordPress with Claude over MCP, you install the official MCP Adapter plugin on WordPress 6.9 or newer, create an Application Password for a dedicated low-privilege user, and add your site’s MCP address to Claude Desktop or Claude Code. As of October 2026 the adapter is in the WordPress.org plugin directory, but it adds no AI features on its own: what Claude can do depends on which abilities your plugins make public.

    This tutorial follows the MCP Adapter’s own documentation and WordPress.org’s docs, checked in October 2026; it does not include our own tests.

    Most of what’s written about the adapter so far assumes you build plugins for a living. This wordpress mcp tutorial is the other version: for site owners and agencies who want to connect Claude to WordPress, run maintenance in plain English, and keep tight control over what the agent is allowed to touch.

    What WordPress MCP actually does

    MCP, the Model Context Protocol, is “an open standard for connecting AI agents to external systems,” in the words of the MCP Adapter’s plugin page. Instead of Claude guessing at your site through a browser, your WordPress install advertises typed capabilities, called abilities, such as “get site info” or “create a draft post.” Claude picks the ability; WordPress checks the connected user’s permissions before anything runs. If the protocol itself is still fuzzy, our plain-English primer on MCP servers for website owners covers it.

    The practical difference from the old REST API workflow is discovery. With the REST API, you (or a developer) had to know every endpoint and wire it up by hand. With the adapter’s default server, Claude sees three tools when it connects: one lists the public abilities on your site, one reads an ability’s details, and one runs it. That is why a non-developer can do a wordpress mcp server setup without writing code. The catch: the adapter “does not provide any AI features itself.” Abilities come from WordPress core or from plugins, only abilities marked public can be reached over MCP, and if none are public yet, the list Claude sees is empty.

    Step 1: Install the MCP adapter

    Requirements first: WordPress 6.9 or newer (the Abilities API the adapter builds on is part of core from 6.9), a site served over HTTPS, and an administrator login. HTTPS matters because WordPress only offers Application Passwords on HTTPS sites, local development sites aside.

    • Update core. Dashboard > Updates. On 6.8 or earlier the adapter has nothing to plug into; the plugin lists WordPress 6.9 as its minimum.
    • Install the adapter. Plugins > Add New, search for “MCP Adapter”, click Install Now, then activate it. With WP-CLI: wp plugin install mcp-adapter --activate.
    • Note your endpoint. Your site’s MCP address is the Site Address (URL) from Settings > General followed by /wp-json/mcp/mcp-adapter-default-server. You’ll paste it into Claude shortly.
    • Add abilities. Install or activate a plugin that registers the abilities you want, or ask the developer of a plugin you already use whether it supports the Abilities API.

    If the address returns a 404 or “rest_no_route,” check that the plugin is active and that your permalinks aren’t set to plain; on plain permalinks, the adapter’s guide says to use ?rest_route=/mcp/mcp-adapter-default-server instead. The same guide notes that some security plugins and hosts turn Application Passwords off, and some Apache setups strip the Authorization header before WordPress sees it.

    Step 2: Create a scoped Application Password

    Don’t connect Claude with your admin account. Create a dedicated user, named something like claude-agent so the audit trail is unmistakable, and give it the lowest role that can do the work. The adapter’s site owner guide suggests “Editor rather than Administrator” and says a separate user “keeps its actions easy to tell apart and easy to switch off.” Then generate an Application Password for that user under Users > Profile > Application Passwords. WordPress shows it only once, so store it in your password manager. The spaces in it are part of the format and can stay.

    This one decision does more for your security than everything else in this guide combined. Every action runs with that user’s role, and per WordPress.org’s roles and capabilities table, an Editor can publish and manage posts and pages, moderate comments and manage categories, but can’t install or update plugins, edit theme files or create users. A badly worded prompt, or a compromised laptop, then has a hard ceiling on the damage it can do.

    Step 3: Connect Claude Desktop or Claude Code

    Both clients work; they connect differently. Claude Desktop reads MCP servers from a JSON file you open under Settings > Developer > Edit Config. The adapter’s guide adds an entry that starts a small helper, @automattic/mcp-wordpress-remote, which runs on your computer and passes Claude’s requests to your site. You need Node.js installed. The entry looks like this, with your own values:

    {
      "mcpServers": {
        "wordpress-site": {
          "command": "npx",
          "args": ["-y", "@automattic/mcp-wordpress-remote@latest"],
          "env": {
            "WP_API_URL": "https://example.com/wp-json/mcp/mcp-adapter-default-server",
            "WP_API_USERNAME": "your-username",
            "WP_API_PASSWORD": "your application password"
          }
        }
      }
    }

    Save the file and restart the app. Claude Code skips the helper and connects over HTTP: one claude mcp add --transport http command with your MCP address and an Authorization: Basic header made from your username and Application Password, joined by a colon and base64-encoded. Copy the exact command from the site owner guide rather than retyping it. Claude Code also runs non-interactively, so the same request can be scripted across several sites. The web version at claude.ai can’t connect on its own: it adds connectors through an OAuth sign-in, and WordPress doesn’t include an OAuth server.

    Either way, run the smoke test the guide recommends: ask Claude to “List the WordPress abilities you can use on my site.” A list, or an answer that the list is empty, means the connection works. An error means it doesn’t. For a 401, check the username (the login name or email address, not the display name), check that the Application Password was copied in full and hasn’t been revoked, and ask your host whether the server passes the Authorization header to PHP.

    Tasks to run first

    Skip the party tricks. Once your plugins expose the matching abilities, these are the kinds of requests worth delegating in the first week. Check the list from the smoke test to see which your site supports today:

    • Draft and stage a post. “Write a short update announcing our new pricing, save it as a draft, and assign the News category.” You review in wp-admin, then publish. Claude drafts; you stay on the button.
    • Content audit. “List every post older than a year that has no internal links pointing to it.” A spreadsheet job becomes a request.
    • Bulk metadata cleanup. “Find posts missing a meta description and draft one for each.” Review the batch, approve, done.
    • Comment triage. “Summarize the pending comments and flag anything that looks like genuine customer feedback rather than spam.”
    • Plain-English queries. “Which categories have had no new posts this quarter?” The agent reads; you decide.

    Notice what is not on the list: nothing destructive, nothing that publishes without review. That is deliberate. For the build side of the same idea, see our guide to whether Claude can build a website.

    Scope what the agent can touch

    The riskiest setup is full admin access on a production site on day one. This containment checklist follows the adapter’s own advice and WordPress’s role model.

    • Least privilege, always. Editor role for content work. Switch to a more powerful role only for a specific task, then switch back the same day.
    • Drafts by default. For the first month, instruct Claude to save everything as a draft. Publishing stays a human click.
    • Staging before production. The adapter’s guide says abilities “change your live site the same way an action in wp-admin does” and recommends trying new abilities on a staging copy first. Point the connection at staging and let it run there before you trust it with production.
    • Revoke without drama. Every Application Password is listed under Users > Profile with when it was last used, and Revoke cuts a client off at once. Create one password per client and device, and revoke the moment a laptop or contractor leaves the picture.
    • Backups before plugin work. If you do grant update rights, take a backup first and update one plugin at a time. Boring beats broken.

    Hosting that won’t fight your agent

    Your host affects how smooth this is. Some hosts and security plugins turn Application Passwords off, some server setups strip the Authorization header, and without staging every experiment happens on your live site. Four options:

    Kinsta pairs well with this setup on paper: it routes traffic through Cloudflare with Enterprise DDoS protection, offers a REST API that Kinsta says is available to all customers while in beta, and starts at $35 a month for one site. Kinsta now operates on Oracle Cloud Infrastructure, which replaced its previous Google Cloud setup. The Kinsta API + Claude automation guide covers that pairing in depth.

    WP Engine lists one-click staging and development environments on every plan, which is exactly where you want an agent living during its probation period, plus EverCache, which its plans page calls “WordPress-optimized caching.” Its entry plan costs $35 a month for one site, or $336 for the first year on annual billing (a new-customer price). If you use its Smart Search AI, which WP Engine says is “Powered by Model Context Protocol (MCP),” our Smart Search setup walkthrough covers it.

    On a budget, Hostinger‘s Premium plan, at $2.99 a month on a 48-month prepay (renewing at $10.99), is a sane place to run your first experiments: spin up a throwaway WordPress install, connect Claude, and break things where it does not matter. Graduate to managed hosting when the agent graduates to production. Our best AI web hosting guide compares the options.

    InstaWP builds this workflow into the host: its InstaMCP feature puts your WordPress site on an MCP server so you can point Claude or Cursor at it, according to InstaWP’s site. Every site starts as a sandbox, $2 a month per site, which is a safe place to test an agent, and moves to managed hosting from $5 a month per site when it goes live, billed by the day, per its pricing page.

    Frequently asked questions

    Do I need to know how to code to use WordPress MCP?

    No. The adapter installs like any plugin. The most technical steps are pasting a JSON entry into Claude Desktop’s config file, with Node.js installed, or running one command in Claude Code. If you can install a caching plugin and follow copy-and-paste instructions, you can do this.

    Is it safe to let an AI agent manage my WordPress site?

    Safe is a spectrum you control. With an Editor account, drafts by default, and a staging site for anything risky, the blast radius is small. With an administrator’s Application Password on production and no backups, it is not safe at all. Every action runs with the connected user’s permissions, so the tooling is neutral; the scoping is the safety.

    What is the difference between the WordPress MCP adapter and the REST API?

    The REST API exposes endpoints that someone has to write integration code for. The MCP Adapter exposes abilities, each with a description and a JSON Schema for its inputs, in a format MCP clients understand, so Claude can discover what’s available the moment it connects. The adapter’s endpoint lives under /wp-json/ alongside the REST API, and both check the connected user’s permissions.

    Can Claude update plugins through MCP?

    Only if two things are true: a plugin on your site exposes a public ability that updates plugins, and the connected user’s role allows it. An Editor can’t update plugins (on a single site that capability belongs to administrators), which is why this guide starts there. When you are ready to delegate updates, grant the capability deliberately, back up first, and have the agent update one plugin at a time.

    Last checked: October 2026. Prices and plans change; check the vendor’s page before you buy.

    Want the guided version? Our free video courses are in production, starting with “Launch Your First Website with Claude.”

  • Can Claude Build a Website? What It Can and Can’t Do (2026)

    Can Claude Build a Website? What It Can and Can’t Do (2026)

    Yes, Claude can build a website, and as of October 2026 there are two ways to do it. claude.ai artifacts, included on every plan from Free up, turn a prompt into a single-page site or landing page that opens beside your chat; Claude Code, included with Pro ($20 a month, or $17 a month billed annually) and higher plans, writes and edits a multi-page project as real files and can run the commands that deploy it.

    This guide is based on Anthropic’s documentation and each host’s published plans, checked in October 2026; it does not include our own tests.

    Neither one hosts the result for the public, and that is the step almost every tutorial skips: getting the finished site onto a real domain instead of leaving it inside Claude. Below: what each tool does well, where each one stops, and the hosting routes that fit.

    The short answer

    Yes. Anthropic’s artifacts guide lists single-page websites among the things Claude can make, and its Design template covers “mockups, prototypes, one-pagers, and landing pages.” For anything bigger, Claude Code works in a real project folder. Either way, you review what it produces: when an artifact throws an error, Anthropic says Claude “will attempt to fix the error, but success isn’t guaranteed.”

    The asterisk: Claude builds websites, it doesn’t host them for the public. Claude itself doesn’t sell hosting or domains, and a shared artifact link only opens for people signed in to Claude. That last mile is where beginners stall, and it’s the part we’ll solve properly in the hosting section below.

    Artifacts vs Claude Code: the two ways Claude builds sites

    Quick verdict first: use claude.ai artifacts to prototype and learn, use Claude Code to ship. Here’s the side-by-side, from Anthropic’s documentation and pricing page.

    claude.ai artifactsClaude Code
    What it isA page, design or small tool that opens in its own window beside your chatAn agentic coding tool that reads your project, edits files, runs commands and works with git
    Best forLanding pages, prototypes, one-page toolsMulti-page sites, WordPress work, anything you’ll maintain
    Skill neededNone: describe what you wantComfort reviewing files and approving commands; runs in a terminal, an IDE, a desktop app or a browser
    Multi-page sitesBuilt for single pages; Design artifacts export as a .zip or standalone HTMLNative: real folders, real files, real links
    PublishingShare links need Pro or higher and open only for people signed in to Claude; export the files to host them yourselfDeploys anywhere you give it access: your host, your domain
    CostIncluded on the Free plan; Pro is $20 a month, or $17 a month billed annuallyPro, Max (from $100 a month), Team or Enterprise plan, or an Anthropic Console account

    Building a landing page in claude.ai

    Start with one detailed prompt: the business, the sections you want (a hero, pricing tiers, a contact section) and the look you’re after. Claude opens the result in its own window beside the conversation. On Free, Pro and Max plans, artifacts need Cloud code execution and file creation turned on under Settings > Capabilities.

    Revisions are where artifacts shine. Ask for changes in plain language, such as a darker hero, a sticky navigation bar or a different pricing order, and the artifact updates as you work. In an artifact started from a template you can also edit directly on the canvas. If something breaks, the “Try fixing with Claude” button sends the error details back to Claude.

    The ceiling shows up fast, though. An artifact is one page. Once you export it, a contact form needs a form service or a server behind it to receive submissions. And sharing isn’t publishing: artifacts start private, and since September 16, 2026, Anthropic’s sharing guide says everyone who opens a shared artifact needs a Claude account (artifacts published from a chat before that date are the exception). For a prototype that’s fine. For a business, export the files (Design artifacts export as a .zip or standalone HTML) and put them on a host.

    Building a real site with Claude Code

    Claude Code is a different animal. Anthropic describes it as “an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools.” For a website, that means it can scaffold a multi-page site with shared navigation and styles, set up version control (Anthropic says it “stages changes, writes commit messages, creates branches, and opens pull requests”), and run the commands that push the result to a server.

    Two capabilities matter most for website work. First, it can debug its own output: for bugs, Anthropic says Claude Code “traces the issue through your codebase, identifies the root cause, and implements a fix.” Second, it handles WordPress. It can run WP-CLI, which WordPress.org says can do “anything you can do in the WordPress admin” from the terminal, or connect to a site over MCP, the open standard Claude Code uses to reach outside tools. What it can do over MCP depends on what the site exposes; we cover how that plumbing works in MCP servers explained for website owners.

    The terminal used to be the main hurdle for beginners. Claude Code now also runs in a desktop app and in the browser at claude.ai/code; Anthropic says the desktop app “includes Claude Code, so you don’t need to install the CLI separately,” though it needs a paid subscription. The payoff is a site you actually own: files in a folder or a Git repository, not a page inside someone else’s app.

    Downsides: what Claude still can’t do

    • No hosting, no domain, no DNS. Claude ends at the code. The internet part is on you (covered below).
    • Generic prompts get generic designs. Push it: reference sites you like, name fonts and colors, or set up a design system, which Anthropic says lets new work pick up “your colors, fonts, and components.”
    • Forms and payments need a backend. Contact forms, bookings and checkout all require a service or a server behind them. Claude can wire these up, but only if you ask and provide the accounts.
    • Check every link and image. Generated code can point to an image file or URL that doesn’t exist. Click through every page before launch.
    • SEO is still your job. Claude can write titles, meta descriptions and markup when you ask, but keyword research and content strategy need your judgment.

    Getting it online: the step every tutorial skips

    A Claude-built site is a folder of files until it lives on a server with your domain pointed at it. Three routes, depending on budget and ambition.

    Budget route: Hostinger is the cheapest of the three. Its Premium plan costs $2.99 a month on a 48-month prepay ($143.52 up front) and renews at $10.99 a month, as of October 2026. Upload Claude’s files with the Hostinger File Manager, a browser tool for uploading, editing and organizing website files. If you’d rather not touch code at all, Hostinger also has its own AI builder; our Hostinger AI builder tutorial walks through it.

    Flexible route: Cloudways runs managed servers on DigitalOcean, Vultr, Linode, AWS and Google Cloud. Its Flexible platform is built for “WordPress, Magento, Laravel & PHP apps” and starts at $11 a month for a 1GB DigitalOcean server, which Cloudways’ getting started FAQ says is “primarily intended for testing and staging purposes”; it advises 2GB or more for live sites, which is $24 a month. You reach your files over SFTP or SSH, so Claude Code can deploy to it, and Copilot, Cloudways’ AI assistant, helps diagnose server problems. For a Node.js app, Cloudways sells a separate product, Cloudways Velocity, from $20 a month.

    Premium WordPress route: if the destination is a serious WordPress site, Kinsta starts at $35 a month for one site, routes traffic through Cloudflare with Enterprise DDoS protection, and offers a REST API that Kinsta says is available to all customers while in beta, which matters when Claude is doing the maintenance. Kinsta is moving its hosting from Google Cloud to Oracle Cloud Infrastructure. Full walkthrough of the deployment side in how to host a Claude Code website, and the wider picks live in our best AI web hosting guide.

    Verdict: who should use which

    If you need a one-page site today, a claude.ai artifact on the Free plan is the quickest start. Just accept that publishing it on your own domain means exporting the files and uploading them to a host yourself. If you’re building something you’ll still be running in a year, start in Claude Code. It needs a paid plan, but the result is a project you own and can version, deploy and hand to someone else.

    And if you’re weighing Claude against the obvious alternative, see Claude vs ChatGPT for building websites.

    Frequently asked questions

    Can Claude build a website for free?

    Yes, a single page. Artifacts are included on the Free plan, and Claude can make single-page websites as artifacts. Usage limits apply on every plan, and Claude Code isn’t included on Free. Hosting the result on your own domain costs extra: Hostinger’s Premium plan, for example, is $2.99 a month on a 48-month prepay.

    Can Claude publish a website directly to the internet?

    Not from a chat. Artifacts start private, and a shared artifact opens only for people signed in to Claude; artifacts published from a chat before September 16, 2026 are the exception. For a public site on your domain, you deploy the files to a host. Claude Code can handle that deployment itself if you give it access to your server or your host’s tools.

    Do I need to know how to code to use Claude Code?

    No. You describe what you want and approve the changes and commands it proposes. The desktop app and the browser version at claude.ai/code avoid the terminal, though both need a paid plan. Reading a little HTML helps you sanity-check the output, and you’ll absorb that as you go.

    Can Claude build a WordPress site?

    Yes, this is where Claude Code earns its keep. It can run WP-CLI to install and update plugins, manage themes and import content, or connect to a live site over MCP, where what it can do depends on what the site exposes. Pair it with a host that offers an API or agent tools and it can help with ongoing maintenance too.

    Is Claude better than a traditional website builder like Wix?

    Different trade. Builders bundle hosting and an editor in one subscription and keep the code out of sight. Claude gives you code you own outright, and you pick the host. If you never want to see a file, a builder is easier. If you want ownership and the freedom to move hosts, Claude is the better fit.

    Last checked: October 2026. Prices and plans change; check the vendor’s page before you buy.

    Want the guided version? Our free video courses are in production, starting with “Launch Your First Website with Claude.”

  • Best AI Web Hosting in 2026: Builders, Assistants, Agents

    Best AI Web Hosting in 2026: Builders, Assistants, Agents

    The best AI web hosting depends on which kind of AI you need. As of October 2026: Hostinger for a budget AI builder ($2.99 a month on a 48-month prepay), 10Web for AI-built WordPress ($10 a month billed annually), Cloudways for an AI server assistant ($11 a month), and Kinsta or WP Engine for premium WordPress ($35 a month). These picks are based on each host’s published plans, documentation and terms, checked in October 2026; they do not include our own tests.

    What actually counts as AI hosting

    Three different things hide under the label, and knowing which one you need decides your host:

    • AI builders: describe your site, get a site. Hostinger and 10Web lead here. You trade some control for speed: Hostinger’s AI Builder has no FTP, SSH or database access.
    • AI assistants: the platform helps you run things, like Cloudways Copilot watching your servers, or Hostinger Agent running WordPress sites by chat (on its Business and Cloud plans).
    • Agent-ready platforms: hosts your own AI agents can operate through APIs and MCP. Hostinger documents an MCP server, Cloudways documents one in its help center, and Kinsta offers a REST API, in public beta, on every plan. Our MCP explainer covers how this works.

    The picks

    Best AI builder on a budget: Hostinger

    Hostinger’s Premium plan costs $2.99 a month on a 48-month term paid up front ($143.52) and renews at $10.99 a month, per its AI Builder page. Hostinger merged its website builder and Horizons into one AI Builder with two modes: Manual for drag-and-drop editing, which uses no AI credits, and Agentic for building by chat, which does (Premium includes 5 as a one-time gift). Its WordPress plans add an AI Website Builder for Elementor or the block editor. Where the AI Builder thins out: no WordPress plugins. Try Hostinger.

    Best for AI-built WordPress: 10Web

    10Web’s Agentic Website Builder creates a WordPress site from a prompt, an existing site or a design file. You edit it in 10Web’s AI Editor or the normal WP Admin, so you aren’t locked into a proprietary builder. AI Starter costs $20 a month, or $10 billed annually, for 1 site and 100 AI credits a month, at 1 credit per AI request. Google Cloud Hosting is listed only on AI Ultimate ($45 a month, or $22.50 annually, for 4 sites). Try 10Web.

    Best AI assistant for cloud servers: Cloudways

    Cloudways runs managed servers on DigitalOcean, Vultr, Linode, AWS or Google Cloud, from $11 a month for 1GB, billed hourly, with unlimited sites per server. Its Copilot assistant monitors servers and applications, explains issues and offers 1-click SmartFix actions, at 1 credit each. A Starter tier with 4 credits comes at no extra cost on qualifying accounts; 12 credits cost $9.99 a month. Try Cloudways.

    Premium WordPress: Kinsta and WP Engine

    When the site is the business, look at these two; both start at $35 a month for 1 site. Kinsta charges $350 a year, with the first month free on select plans and a 30-day money-back guarantee. Its REST API can create sites, push staging to live, update plugins and restore backups, and Kinsta is moving its hosting from Google Cloud to Oracle Cloud Infrastructure. WP Engine’s first year costs $336 for new customers, and renewal pricing is subject to change. Every WP Engine site includes staging and development environments; Smart Search AI costs extra, from $140 a month. Our full head-to-head compares them by use case. Try Kinsta · Try WP Engine.

    Which one for your situation

    • “I need a site this weekend and I’m not technical”: Hostinger’s AI Builder.
    • “I want AI to build it, but in real WordPress”: 10Web, or Hostinger’s WordPress plans.
    • “I’ve outgrown shared hosting but don’t want to be a sysadmin”: Cloudways.
    • “This site makes money and downtime costs me”: Kinsta or WP Engine.
    • “My AI agents should build and run the sites”: Hostinger or Cloudways, which both document MCP servers, or Kinsta if you’d rather script a REST API. See our agent hosting guide.

    The picks at a glance

    HostFromAI categoryStandoutSkip if
    Hostinger$2.99/mo (48-month prepay)Builder / agent-readyAI Builder plus an MCP serverYou need plugins, FTP or SSH on the AI Builder
    10Web$10/mo billed annuallyBuilderAI output is editable WordPressYou want infrastructure control
    Cloudways$11/mo, billed hourlyAssistant / agent-readyCopilot AI plus an MCP serverYou never want to think about servers
    Kinsta$35/mo or $350/yrAgent-friendly APIREST API on every planBudget is the constraint
    WP Engine$35/mo ($336 first year)Assistant (paid add-on)Staging and dev environments on every siteYou want a published renewal price

    What about Bluehost, GoDaddy, and the other big names?

    They’re not here because we haven’t checked their current AI features against their documentation, not because of a verdict on them. If you’re weighing one, use the same test: is its AI a builder, an assistant, or something your own agents can drive through an API or MCP server? Is it included or an add-on? Then read the renewal price as closely as the feature list.

    What AI builders produce, according to their docs

    Both builders meter the AI work. Hostinger says a simple business page usually takes 5 to 10 credits in Agentic mode, so Premium’s 5 one-time credits won’t go far without a top-up. 10Web charges 1 credit per AI request against 100 a month on AI Starter. Treat any first draft as a skeleton: rewrite the copy and replace images. The bigger difference is how editable the result is afterward. 10Web’s output is a WordPress site with backups you can download. Hostinger’s exports are partial: Manual mode exports pages, text and images to WordPress, but not styles, layout, the store, forms or SEO settings, and Agentic mode exports a React and Vite project without your backend data.

    The pricing traps, named

    • The renewal cliff. Hostinger’s $2.99 headline needs a 48-month prepay and renews at $10.99 a month, about 3.7 times the intro rate. WP Engine’s $336 first year is a new-customer price. Do the year-two math before you commit.
    • Visit-based overages. Managed WordPress plans meter traffic: Kinsta charges $0.50 per 1,000 visits over the cap, WP Engine $2 per 1,000 and 10Web $2 per 10,000. Know your monthly visits before picking a tier.
    • AI-credit meters. Hostinger’s Agentic mode, 10Web’s AI Editor and Cloudways Copilot all spend credits. Check the allowance and how top-ups work before heavy iteration.
    • The migration hostage. Any host worth using helps you leave. If exporting your site looks deliberately painful in the docs, believe the docs and go elsewhere.

    How we chose

    Every fact here comes from the hosts’ own pages, checked in October 2026. We weight the AI features a host documents (builders, assistants, APIs and MCP servers) over marketing checklists, then what the entry plan includes, then price honesty: an intro rate means little if renewal more than triples it. If the documented reason for a pick changes, so does the pick.

    Where to go next

    A worked example. Say you run a consultancy site at about 30,000 visits a month. Start the shortlist at Cloudways vs Kinsta: Kinsta’s $35 Single plan covers 35,000 visits a month with no server to size, while Cloudways starts at $11 a month for a 1GB server with unlimited visits, and sizing it is your job. Add a store and the math changes. Kinsta recommends plans with 4 or more PHP threads for ecommerce, and its cheapest Single plan with 4 is Single 65k at $50 a month. Our PHP workers explainer and the Liquid Web vs Kinsta matchup cover why that matters for stores.

    Budget-first readers should walk the upgrade path instead: Hostinger vs Cloudways lists the signs it’s time to move up from a $2.99 plan. And if you want the whole field in one table, the managed WordPress roundup compares these five hosts plus Liquid Web by scenario, from budget sites to WooCommerce and agencies.

    Frequently asked questions

    Is AI web hosting worth paying extra for?

    Usually you don’t have to: most AI features covered here come with normal plans. Hostinger’s $2.99 Premium plan includes its AI Builder, Cloudways includes 4 Copilot credits on qualifying accounts, and 10Web’s plans include monthly AI credits. The exception is WP Engine’s Smart Search AI, from $140 a month. Pay for hosting quality first; treat AI features as the tiebreaker.

    Can an AI really build a whole website?

    It can build a structured first draft from a prompt, and 10Web can also start from an existing site, a screenshot or a Figma design. Expect to refine copy, swap images and adjust layout, and budget AI credits for those edits.

    What’s the difference between an AI builder and an AI agent managing my site?

    A builder is a one-time wizard: it generates the site, then you’re on your own. An agent is ongoing staff: connected through MCP or an API, it keeps building, updating and maintaining over time. Hostinger and Cloudways already document MCP servers for that kind of work.

    Can I move my site if I pick wrong?

    Moving in is the easy part: Kinsta includes unlimited free migrations, Hostinger and Cloudways advertise free migration, WP Engine includes an automated migration plugin, and 10Web lists automated 1-click migration. Moving out depends on what you built. WordPress sites are portable by design; Hostinger’s AI Builder exports only part of a project, so read the export docs before you start.

    Do these recommendations change?

    Yes. This page reflects each host’s published plans, prices and documentation as checked in October 2026, and the list changes when they do.

    Last checked: October 2026. Prices and plans change; check the vendor’s page before you buy.

  • How to Set Up a VPS for Your AI Agents (Step by Step)

    How to Set Up a VPS for Your AI Agents (Step by Step)

    This is the practical companion to our AI agent hosting guide: a start-to-finish walkthrough for giving an agent its own server. Thirty minutes, $5–20 a month, and you never have to worry about an agent experiment eating your laptop again.

    Step 1 — Pick a provider and size

    Any major VPS provider works: Hetzner, DigitalOcean, Vultr, Linode, and friends all offer the two things that matter — root SSH and an API. For a single agent doing web work, 2 vCPU / 4GB RAM / 40GB disk is a comfortable start; you can resize later, so err small. Choose the datacenter region closest to the services your agent talks to, not to you — the agent doesn’t care about its own ping times to your house.

    Step 2 — Provision with a key, not a password

    Create the server with Ubuntu LTS (the boring choice is the right choice — every tool documents Ubuntu first). Add an SSH public key at creation time and disable password login from day one. If your agent framework manages its own keys, generate a dedicated keypair for the agent rather than reusing yours — you want to be able to revoke the agent without locking yourself out.

    Step 3 — The ten-minute hardening pass

    • Create a non-root user for the agent with sudo where needed: adduser agent && usermod -aG sudo agent
    • Firewall: ufw allow OpenSSH && ufw enable, then open only the ports the agent’s services actually need.
    • Unattended security updates: apt install unattended-upgrades.
    • fail2ban for SSH: apt install fail2ban. Done.

    Step 4 — Install the agent’s toolbox

    Most agents want the same base kit: git, docker, a modern runtime (Node and/or Python), and whatever CLI your agent framework uses. Docker earns its place immediately — it lets the agent run databases, apps, and experiments in containers it can destroy without hurting the box.

    Step 5 — Connect your agent

    How the agent reaches the server depends on your stack, but the patterns are consistent: SSH access configured with the agent’s own key; environment variables or a secrets file for API tokens (never hard-coded); and if the agent manages applications like WordPress, an MCP connector or application password scoped to that one site — our MCP explainer covers why that beats screen-driving an admin panel.

    Step 6 — Snapshot before you let it loose

    Take a provider snapshot the moment the box is configured and clean. That snapshot is your undo button for everything the agent does afterward. Snapshot again before any big experiment. Storage is pennies; rebuilding a server from memory is an afternoon.

    The whole setup as one copy-paste block

    Once you’ve done steps 2–4 manually and understand them, here’s the condensed version for every server after the first. Run as root on a fresh Ubuntu box, then log out and back in as the agent user:

    adduser --disabled-password --gecos "" agent
    usermod -aG sudo agent
    mkdir -p /home/agent/.ssh
    cp ~/.ssh/authorized_keys /home/agent/.ssh/
    chown -R agent:agent /home/agent/.ssh
    ufw allow OpenSSH && ufw --force enable
    apt update && apt -y install unattended-upgrades fail2ban git docker.io
    usermod -aG docker agent
    systemctl enable --now docker fail2ban

    Nine lines, ninety seconds, and the box matches everything this guide covered. Better yet: paste this guide’s URL into your agent and let it run the setup itself — that’s not a joke, it’s the workflow.

    Knowing it’s alive: minimum viable monitoring

    You don’t need an observability stack for one agent box. You need two things: a free uptime ping (UptimeRobot, Better Stack, or your provider’s built-in monitoring) pointed at any service the agent exposes, and the provider’s billing alert set at twice your expected spend. The first tells you the box died; the second tells you the agent got creative. Add real monitoring when you add real workloads — not before.

    When something breaks: the five-minute triage

    Sooner or later the agent reports it can’t reach the server, or the server stops answering entirely. Before you rebuild anything, walk this list — it resolves the majority of incidents:

    • Can you SSH in yourself? If yes, the server is fine and the agent’s credentials or config drifted. Check its key and its known_hosts.
    • Provider console says what? Every VPS provider has an emergency console that works even when SSH doesn’t. If the box is out of memory, you’ll see it here first.
    • Disk full? The most common agent-inflicted wound — logs and Docker images pile up. df -h then docker system prune fixes more incidents than any other command.
    • Did the agent change the firewall? If it was “improving security” recently, it may have locked you both out. The provider console gets you back in.
    • Still stuck after ten minutes? Restore the snapshot. This is why you took it. A rebuilt-from-snapshot server in four minutes beats an archaeology session every time.

    Notice what’s not on the list: panic. Nothing on a properly set up agent box is unique or irreplaceable, so the worst case is always “restore and move on.”

    The rules that keep this safe

    • One agent, one server (or one container) — isolation is the whole point.
    • The agent’s box holds nothing you’d cry about losing. Backups flow off the box, not onto it.
    • Scoped, revocable credentials only. When in doubt, make a new token.
    • Set a billing alert at 2× expected spend. Loops happen.

    That’s the whole setup. The free Hosting for AI Agents course will go deeper on multi-agent fleets, MCP servers, and production patterns — this page gets your first agent its first home today.

    Frequently asked questions

    Which VPS provider is best for AI agents?

    Any provider with root SSH, an API, and snapshots works — the workflow in this guide is identical across Hetzner, DigitalOcean, Vultr, and Linode. Pick on price and datacenter region. Hetzner tends to win on price per GB of RAM; DigitalOcean has the gentlest interface for first-timers.

    How big a server does an agent need?

    Start with 2 vCPU and 4GB RAM. That comfortably runs a web-work agent with Docker. Resizing up later is a five-minute operation at every major provider, so buying headroom in advance mostly wastes money.

    Should each agent get its own server?

    One agent per server is the simplest safe pattern and the right starting point. Move to one bigger box with a container per agent only when the fleet is large enough that the per-server cost hurts — and accept that isolation gets a little weaker when you do.

    Is it safe to give an agent root access?

    On its own disposable box, with nothing valuable stored there and a snapshot to roll back to — yes, that is the design. The rule that matters is the blast radius: the agent can have root on its server precisely because that server holds nothing you cannot rebuild in ten minutes.

  • MCP Servers Explained for Website Owners

    MCP Servers Explained for Website Owners

    If you run a website and you’re starting to use AI agents, you’ll keep hitting one question: how does the agent actually manage my site? The old answer was screen-scraping admin panels or bespoke API glue. The emerging answer is MCP — the Model Context Protocol — and it’s worth understanding even if you never write a line of code.

    MCP in one paragraph

    MCP is an open protocol that lets AI assistants connect to tools and data sources through a standard interface. Instead of an agent pretending to be a human — clicking buttons, filling forms, breaking every time a UI changes — the tool publishes a menu of capabilities (“create a post,” “list plugins,” “run a query”), and the agent calls them directly. Think of it as USB for AI: one connector standard, many devices.

    What this means for your website

    With an MCP server attached to your site, an agent can create and edit content, manage plugins and themes, run maintenance, and diagnose problems — through clean, permissioned operations rather than a browser puppet show. For WordPress specifically, MCP plugins expose the whole site as a set of tools: content, media, database, files, even site logs. The practical difference is reliability: a UI click-bot breaks when a button moves; a protocol call doesn’t care what the admin screen looks like.

    The security questions to ask first

    • Scope: Can you grant read-only vs. admin capability per connection? You want tiers, not all-or-nothing.
    • Revocation: Is access a token you can kill instantly, separate from your own login?
    • Dangerous operations: Are code execution and database writes off by default, opt-in per site? They should be.
    • Audit trail: Can you see what the agent did? Logs turn “something broke” into “this call broke it.”

    How it works, gently technical

    An MCP server is a small program that sits in front of a system and publishes three kinds of things: tools (actions the agent can take — “create a post,” “list plugins”), resources (data it can read), and prompts (reusable templates for common tasks), each with a description in a format the AI understands natively. When your agent connects, it downloads that menu, and from then on “add a testimonial to the about page” becomes a structured tool call with typed parameters — not a guess about which button to click.

    The connection itself is authenticated with a token, which is where the practical security lives: the token has a scope (what tools it may call) and a kill switch (revoke it, and the agent is out instantly, with your own login untouched). On a WordPress site, a good MCP plugin adds one more layer — site-level toggles for the dangerous capabilities, so code execution and database writes are off until an administrator turns them on for that specific site.

    What to look for in a WordPress MCP plugin

    The ecosystem is young and quality varies. When you evaluate one, weight these over feature-count:

    • Granular capability toggles. Content editing, plugin management, file access, SQL, and code execution should each be separately switchable — and the risky ones off by default.
    • Safety rails on the dangerous tools. The best plugins syntax-check code before writing it and can auto-roll-back a change that takes the site down. That single feature has saved this site more than once.
    • Block-editor awareness. A plugin that understands Gutenberg blocks produces content that survives the editor. One that just writes raw HTML leaves landmines for the next human editor.
    • Activity logging. Every tool call, timestamped. Non-negotiable for the day something looks wrong.
    • Active maintenance. MCP is evolving fast; a plugin that hasn’t shipped in six months is already behind the protocol.

    Full disclosure of method: this entire site — pages, styling, the post you’re reading — is built and maintained through exactly this kind of connection. The evaluation criteria above aren’t theoretical; they’re what we depend on daily.

    MCP vs. plain APIs vs. browser automation

    Three ways an agent can operate your site, and where each one belongs:

    • Browser automation (the agent drives a real browser): the last resort. It works on anything with a screen, and it breaks on anything with a redesign. Reserve it for services that offer no other door.
    • Plain REST APIs: reliable and fast, but every service speaks its own dialect, so each integration is custom work. Great when you’re wiring one specific thing.
    • MCP: the standardized layer on top. The agent learns one protocol and every MCP server — your site, your database, your project tracker — presents itself the same way. Less glue code, and the tool descriptions travel with the connection, so the agent knows what it can do without you explaining.

    In practice they stack: MCP where it exists, raw API where it doesn’t, browser automation when there’s no other way in. A well-run site pushes as much as possible into the first column — this one runs on it daily.

    Where this is heading

    Every serious platform is growing a machine interface, and hosting providers are starting to advertise MCP support the way they once advertised one-click WordPress installs. Sites that agents can manage cleanly will get maintained more, faster, and cheaper than sites that require a human in a control panel for every change. When you evaluate a host or a platform from here on, add one question to your checklist: “how does an agent operate this?” If the answer is a shrug, that platform is aging out.

    Next steps: the AI agent hosting guide for the infrastructure picture, and the VPS walkthrough when you’re ready to give your agent a home of its own.

    Frequently asked questions

    Is MCP only for WordPress?

    No — MCP is platform-neutral. WordPress happens to have mature MCP plugins, but MCP servers exist for databases, file systems, project tools, browsers, and hundreds of SaaS products. The protocol is the same everywhere, which is exactly the point.

    Does MCP replace my site’s admin panel?

    It sits alongside it. You keep the admin panel for yourself; the agent gets the protocol. Both operate the same site — the difference is that the agent’s path is permissioned, logged, and does not break when a menu moves.

    Is it safe to connect an AI agent to my website?

    With scoped access, yes — grant read-only or content-only capability first, keep code execution off until you need it, and use a token you can revoke instantly. The checklist in this guide covers the four questions to ask before connecting anything.

    What happens if the agent makes a bad change?

    The same thing that happens when a human does: you restore. Keep backups or snapshots current, prefer platforms with revision history (WordPress keeps one for content), and review the agent’s log so you know what changed. Boring safeguards beat clever ones.

  • What Is AI Agent Hosting? The 2026 Guide

    What Is AI Agent Hosting? The 2026 Guide

    Your AI agents can already write code, manage your calendar, and answer your email. The next thing they need is the same thing every capable employee eventually needs: their own infrastructure. This guide explains what AI agent hosting actually means, why your existing shared hosting won’t cut it, and how to choose a setup that works — whether you’re running one assistant or a fleet.

    What is AI agent hosting?

    AI agent hosting is server infrastructure provisioned for software agents to use — not just to serve pages to humans. That distinction changes the requirements. A human visitor loads a page and leaves. An agent connects over SSH or an API, runs commands, writes files, installs packages, manages long-running processes, and comes back every few minutes to check on its work.

    In practice, “hosting for AI agents” covers three overlapping needs:

    • A workspace — somewhere an agent can execute code, store files, and run services without touching your laptop. Usually a VPS or container.
    • Managed surfaces — websites, databases, and apps the agent administers on your behalf, through APIs or protocols like MCP rather than by clicking around an admin panel.
    • Isolation — boundaries that keep an agent’s mistakes (and every agent makes them) contained to its own sandbox instead of your production environment.

    Why shared hosting fails agents

    Traditional shared hosting was designed around a human with an FTP client and a control panel. Agents break its assumptions almost immediately: no root access to install the tools they need, no long-running processes allowed, aggressive resource limits that kill jobs mid-task, and CAPTCHAs or panel UIs that automation can’t (and shouldn’t) drive. If a host’s answer to “how do I automate this?” is “log into cPanel,” it isn’t agent-ready.

    What agent-ready infrastructure looks like

    • Root SSH access — the baseline. If your agent can’t open a shell, it can’t work.
    • An API for provisioning — creating, resizing, snapshotting, and destroying servers programmatically.
    • Predictable billing — flat monthly VPS pricing beats per-request serverless when an agent might loop. You want a cost ceiling, not a surprise.
    • Snapshots and rollback — agents experiment. One-command restore turns a bad experiment into a shrug.
    • MCP or API access to the apps themselves — for WordPress and similar platforms, a proper machine interface beats screen-driving the admin every time. (More in our MCP explainer.)

    The three common setups

    1. One VPS per agent (the default)

    A small cloud server ($5–20/month) dedicated to a single agent or project. Simple mental model, clean isolation, easy to snapshot and destroy. This is where most people should start — our step-by-step VPS guide walks through it.

    2. One bigger box, containerized

    A mid-size server running Docker, one container per agent workload. Cheaper at fleet scale and faster to spin up, at the cost of slightly weaker isolation and a bit more ops knowledge.

    3. Managed platforms with agent interfaces

    Hosts that expose sites and servers through APIs and MCP connectors, so agents manage the application while the platform owns the underlying server. The fastest path when your agents’ job is running websites rather than arbitrary code.

    Security: the part you don’t skip

    Give an agent the narrowest credentials that let it do its job. Scoped API tokens instead of your account password. Application passwords you can revoke instead of your admin login. A firewall that assumes the agent’s box may misbehave. And keep agents off machines that hold things they don’t need — your agent’s server should know nothing about your banking.

    How to choose between the three setups

    The decision usually makes itself once you answer two questions: how many agents, and what do they do?

    • One or two agents doing real work → one VPS each. The isolation is worth more than the $10 you’d save consolidating, and you can destroy a misbehaving box without a second thought.
    • Five or more agents, or lots of short-lived experiments → one bigger box with Docker. You’ll trade some isolation for much faster spin-up and roughly half the cost, and by the time you’re running five agents you know enough Docker to be safe about it.
    • Agents whose whole job is managing websites → a managed platform with MCP or API access. There’s no reason to babysit an operating system when the platform will do it, and the agent gets a cleaner interface than raw SSH anyway.

    Mixing models is normal. Around here, sites live on a managed WordPress platform the agents reach over MCP, while scratch work and pipelines get disposable VPSes. Use the boring option for each job.

    What it actually costs

    Real numbers, mid-2026, no affiliate spin:

    • Single agent VPS: $5–12/month gets 2 vCPU / 4GB at the budget providers; $18–24 at the premium ones. Backups add ~20%.
    • Small fleet on one box: a $40–60/month 8GB server comfortably runs 4–8 containerized agents that aren’t doing heavy compute.
    • Managed WordPress with agent access: $10–30/month per site depending on tier — you’re paying for the platform to handle updates, backups, and the machine interface.
    • The hidden line item: LLM API usage usually dwarfs hosting. A $10 server directing $200 of monthly model calls is the normal shape of the bill, so optimize prompts before you optimize servers.

    The agent-ready checklist (use it on any host)

    Evaluating a provider? Run down this list before you hand over a card number. Every “no” is future friction:

    • Root SSH on every plan, not just the expensive ones?
    • A documented API that covers create, resize, snapshot, restore, and destroy?
    • Snapshots priced by storage (pennies), not by count?
    • Flat monthly pricing with a visible cap — or at minimum, spending alerts?
    • Can you add a second SSH key without support tickets?
    • Does account access support scoped API tokens, so the agent never holds your master login?
    • If it’s a managed platform: is there an MCP connector or real API for the application layer, and can dangerous operations be toggled per site?

    Seven questions, five minutes, and it filters the market brutally well. Most budget VPS providers pass. Most legacy shared hosts fail on the first line and it only gets worse from there.

    The bottom line

    Agent hosting isn’t exotic — it’s ordinary infrastructure chosen with a machine operator in mind: root access, APIs, isolation, rollback, flat pricing. Get those five right and your agents become genuinely useful employees instead of toys that live and die inside a chat window. Start with the VPS setup guide, and check the free Hosting for AI Agents course when it opens.

    Frequently asked questions

    Can I run AI agents on my own computer instead?

    For development and experiments, absolutely — that’s where most people start, and it costs nothing. The limits show up fast though: the agent stops when your machine sleeps, your home IP gets rate-limited or blocked by services that see automation, and one runaway process can make your work computer miserable. The usual path is local for the first week, then a cheap VPS the first time an overnight job matters.

    What about serverless or edge functions for agents?

    Good for short, event-driven tasks — a webhook that triggers a five-second job. Wrong for the persistent, stateful work most agents do: execution time limits kill long tasks, cold starts add friction, and per-invocation pricing is exactly the billing model you don’t want attached to software that can loop. Agents want a boring computer that’s always there.

    Do AI agents really need their own server?

    For anything beyond toy tasks, yes. An agent that runs on your laptop stops working when the lid closes, and an agent sharing your production server can take your site down with an honest mistake. A $10/month VPS removes both problems, which is cheap insurance for something that works while you sleep.

    How much does AI agent hosting cost?

    A single-agent VPS runs $5–20/month at providers like Hetzner, DigitalOcean, or Vultr. A containerized fleet on one mid-size box lands around $40–80/month for several agents. The bigger cost risk is metered serverless pricing when an agent loops — flat monthly pricing is the safer default.

    Can I host AI agents on shared hosting?

    Mostly no. Shared hosting blocks root access, kills long-running processes, and assumes a human with a browser. Some agent tasks that only need an API (like managing WordPress through MCP) work fine against a shared-hosted site — but the agent itself should live elsewhere.

    What is the difference between AI agent hosting and regular VPS hosting?

    The hardware is identical — the difference is configuration and discipline: key-only SSH for a non-human operator, scoped credentials you can revoke, snapshots before experiments, and a firewall that assumes the tenant may misbehave. Same server, different tenant.