2026-05-17 · KY · ~5 min read

How to Connect Your Website to Claude AI Using MCP (No Coding Required)

Claude is great at general knowledge and terrible at your website. Out of the box, Claude either guesses based on training data months out of date or fetches one page on demand and pretends to understand the rest. This post walks through the five-minute fix: connect your website as a Model Context Protocol (MCP) server so Claude can answer about every page, with citations.

What you'll have at the end

Claude Desktop with a new connected server in its sidebar. Asking it any question about your site — pricing, refund policy, the API auth flow, the latest changelog entries — returns an answer that cites real source URLs from your pages, not guesses.

Total time: about five minutes. No coding. No self-hosted server. Free tier covers personal docs and evaluation use cases.

What you need before starting

Claude Desktop installed — download from Anthropic if you haven't.
A public website URL — your docs, your blog, your store, your wiki. Anything Claude needs to know about.
A Google account to sign in to WebToMCP (free tier, no credit card).

That's it. No package install (unless you prefer the CLI path). No config files unless you want them.

Step 1 — Index your website

Sign in to webtomcp.net with Google, click New knowledge base, paste your website URL. WebToMCP crawls the site — respecting robots.txt and the path prefix you submit — chunks each page, embeds the chunks into a hybrid vector + keyword search index on Cloudflare's edge, and surfaces progress live on the dashboard.

Most sites under 500 pages finish in 1–10 minutes. JavaScript-rendered sites (SPA dashboards, e-commerce stores with client-side rendering) take longer and are metered per plan. The Free tier covers 5 JS-rendered pages per month; Hobby ($9/mo) covers 100.

Tip — path-prefix scoping. Submit https://docs.example.com/getting-started and we only crawl URLs under /getting-started. Want the whole docs site? Submit the docs root. Each knowledge base stays focused.

Step 2 — Copy the MCP endpoint URL

When the knowledge base flips to Ready, click into it. The detail page shows your MCP endpoint URL:

https://mcp.webtomcp.net/batches/<your-batch-id>/mcp?key=wmcp_<your-key>

Click Copy. The URL embeds both the batch id and an access token (?key=) — treat it like a personal access token (don't post it on a public page).

Step 3 — Add it to Claude Desktop

Two paths. The CLI is faster; the manual config is more transparent.

Path A — CLI (one command)

npm i -g webtomcp-cli
wmcp connect claude

The CLI walks you through pasting the endpoint URL, writes the entry into Claude Desktop's config file, and prompts you to restart Claude. Done.

Path B — Manual config edit

If you'd rather edit the config directly, open Claude Desktop's config.json:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

Add (or merge) under mcpServers:

{
  "mcpServers": {
    "webtomcp-docs": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.webtomcp.net/batches/YOUR-BATCH-ID/mcp?key=wmcp_YOUR-KEY"
      ]
    }
  }
}

Replace the URL with the one you copied in Step 2, save the file, then quit Claude Desktop fully (Cmd-Q on macOS, right-click tray → Quit on Windows) and reopen. Closing the window alone is not enough — Claude only reads the config at startup.

Step 4 — Verify it works

In Claude Desktop, click the + or hammer icon — you should see your new server (webtomcp-docs or whatever name you chose) listed. Ask Claude a question only the indexed site would know:

"What's the refund policy on this docs site?"

Claude should return an answer with source URLs from your indexed pages. If it does, you're done.

What Claude can do once connected

The connected MCP server lets Claude answer anything a colleague would know after reading every page of your site:

"Compare the Pro and Business pricing tiers."
"Summarize the API authentication flow, with code samples."
"Anything about webhooks in the changelog this month?"
"Onboarding steps for a new customer in the EU."
"Walk me through how to deploy to staging."

Every response cites the source URL on your site, so you can click through to verify. Claude no longer hallucinates pricing tier names, invents API endpoints, or quotes outdated policy.

Troubleshooting

Claude doesn't show the server after restart. Quit Claude Desktop fully — Cmd-Q on macOS or right-click tray → Quit on Windows. Closing the window isn't enough; the config is only read at startup.

"Failed to connect to MCP server". Double-check the URL has both the batch id and the ?key=... query string. Re-copy from the WebToMCP dashboard if unsure.

Crawler stuck at 0 pages. Some sites block crawlers in robots.txt, return Cloudflare bot challenges, or 401 unauthenticated requests. The WebToMCP dashboard surfaces the specific error inline. For JS-heavy sites, flip Enable JS rendering on the batch.

Answers feel stale. WebToMCP re-crawls weekly by default. Click Re-crawl in the dashboard for an instant refresh.

When this is worth doing (and when it isn't)

Connecting your website as an MCP server pays off when:

Your audience already uses Claude for their day-to-day work and would benefit from accurate, cited answers about your product or content.
Your site has more than a handful of pages (docs, knowledge bases, changelogs, wikis, product catalogs).
You'd rather an AI cite real pages than hallucinate from a six-month-old training snapshot.

It's overkill for a five-page brochure site or a single landing page. It's high-leverage for any developer-facing docs site, customer-facing knowledge base, or content-heavy e-commerce catalog.

Questions? developer@webtomcp.net. Or sign in with Google to index your own site (free).