How to query documentation directly from Cursor using MCP

Cursor's AI can write code, explain errors, and refactor — but it loses the thread the moment you need to reference a library's docs. You break out of the flow, open a browser tab, find the page, come back. This is fixable. Here's how to set up Cursor so you can query any documentation without leaving your editor.

How it works

Cursor supports MCP — the Model Context Protocol. MCP lets Cursor's AI call external tools during a conversation. AgentReady provides an MCP server that indexes websites and exposes them as queryable tools. Once connected, you can ask Cursor's AI "what does the Tailwind docs say about container queries?" and it'll look it up in indexed content and respond with the actual answer.

Setting up AgentReady in Cursor

Open Cursor settings (Cmd+, on macOS) and navigate to Features → MCP Servers. Click Add new MCP server and enter:

{
  "agentready": {
    "command": "npx",
    "args": ["-y", "@agentreadyweb/mcp"]
  }
}

Alternatively, edit your ~/.cursor/mcp.json directly and add the same block.

Restart Cursor. You should see the MCP server listed as connected in the MCP settings panel.

Index the docs you want to query

Before querying, the site needs to be indexed. You can do this:

Via AgentReady portal — go to agentready.it.com, paste the docs URL, wait 60 seconds.

Via Cursor chat — in Agent mode, ask "use AgentReady to index tailwindcss.com/docs". Cursor will call submit_site automatically.

Automatically — when you ask about a site that isn't indexed yet, ask_site will index it first and then answer. The first query takes longer (~60s) but it works without any explicit submit step.

Querying docs in practice

With AgentReady connected, try these prompts in Cursor Agent mode:

"Check the React docs and tell me how to use useCallback correctly"
"What does the Prisma docs say about transactions?"
"How does Vercel handle environment variables according to their docs?"
"Ask AgentReady about Stripe webhook signature verification"

Cursor's agent will call ask_site with the domain and question, receive indexed content, and generate a response grounded in the actual documentation — not its training data, which might be out of date.

Why this beats copy-pasting from docs

When you paste a docs URL into Cursor's context, it fetches that page's raw HTML, strips it roughly, and works with whatever text it extracts. That's one page — and if the answer is on a different page, you don't get it.

AgentReady indexes up to 10 pages of a site and does proper semantic chunking. When you ask a question, it retrieves the most relevant chunks across all indexed pages — not just the one you happened to paste. If the pagination answer is on the "advanced queries" page and the authentication answer is on a different page, both are reachable.

Checking what's already indexed

Tell Cursor to list available sites: "what sites does AgentReady have indexed?" — it'll call list_sites and show you what's available. The directory page also shows all indexed sites.