How to set up MCP in Claude Code — query any docs from your terminal

Claude Code — Anthropic's terminal-based AI coding assistant — supports MCP servers natively. Once you add AgentReady's MCP server, you can ask Claude Code to look up any indexed documentation without leaving your terminal: the right answer, drawn from the actual source, with citations.

This guide covers the setup, how to verify it's working, and what to ask once it's connected.

Prerequisites

• Claude Code installed (npm install -g @anthropic-ai/claude-code)
• Node.js 18+ on your PATH
• That's it — no API key for AgentReady, no account required for the first site

Step 1 — Add the AgentReady MCP server

Run this from anywhere in your terminal:

claude mcp add agentready npx @agentreadyweb/mcp

This registers the server globally. Claude Code will start it automatically whenever it needs to use an AgentReady tool. You can verify it was added with:

claude mcp list

You should see agentready in the output with status configured.

Step 2 — Index the docs you want to query

Any public site can be indexed. You can do it directly from Claude Code:

# In a Claude Code session (run: claude)
> Use AgentReady to index supabase.com/docs

Claude Code will call the submit_sitetool, wait for indexing (~60 seconds), and confirm when it's done.

Alternatively, index from the web at agentready.it.com and the site will be immediately available to any MCP client including Claude Code.

Many sites are already indexed. Check with:

> What sites does AgentReady have indexed?

Step 3 — Query in your workflow

Start a Claude Code session in your project directory and ask questions naturally:

# Start Claude Code
claude

# Then ask anything about indexed docs
> What does the Supabase docs say about Row Level Security?
> Check the Stripe docs — how do I handle webhook signature verification?
> What's the Tailwind CSS way to add dark mode?
> According to the Prisma docs, how do I use transactions?

Claude Code will call ask_site with the relevant domain and question, retrieve indexed content, and respond with a cited answer. The response comes from the actual documentation — not training data that might be outdated.

How it's different from web_fetch in Claude Code

Claude Code already has a WebFetchTool that can fetch URLs. The difference matters in practice:

• web_fetch fetches one page— if the answer is on a different page, you don't get it. You have to know the exact URL and hope the answer is on that page.
• AgentReady searches across all indexed pages— it retrieves the most relevant chunks from up to dozens of pages based on your question. The authentication answer and the pagination answer are both reachable even if they're on different pages.
• JavaScript SPAs — React, Next.js, and Vue apps return empty HTML on a raw fetch. AgentReady uses headless rendering during indexing, so the content is there regardless of how the site is built.

Project-scoped vs global MCP config

The claude mcp add command registers the server globally (in ~/.claude/settings.json). If you want to scope it to a specific project or share it with your team via version control, create a .claude/settings.json in the project root:

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

Commit this file and every developer on the project gets AgentReady's tools in their Claude Code sessions automatically.

Troubleshooting

Server not starting: confirm Node 18+ is on your PATH with node --version. npx needs network access to download the package on first run.

Tools not appearing: run claude mcp listto confirm registration. If the server shows as configured but tools aren't used, try explicitly asking Claude Code: “use AgentReady to answer this” — it will call the tool rather than relying on training data.

Site not indexed: if you ask about a domain and Claude Code says it isn't indexed, ask it to submit the site: “use AgentReady to index prisma.io”. The first query after indexing takes ~60 seconds.