Reader MCP Server

Serve your documentation to AI tools

The Reader MCP Server is your own MCP endpoint, served from your documentation domain. It gives MCP-compatible tools read-only access to your published documentation so they can search, retrieve, and reference your content without any ability to change it. Public documentation is available without authentication. Private, partial, and password-protected documentation is accessible through two authentication types: OAuth 2.1 for interactive client connections, and machine-to-machine credentials for automated access without a browser.

This is the server your readers connect to: customers, support teams, solution engineers, and internal users, along with the AI tools they work in. If you are looking for the server your documentation team connects to in order to write docs, see the Authoring MCP Server.

Find your endpoint

The Reader MCP Server becomes available as soon as you publish your documentation. There is nothing to provision or enable.

You can find your endpoint in the dashboard under MCP Servers in the sidebar. It follows the same shape as your documentation URL:

• https://docs.yourdomain.com/_mcp
• https://yourdomain.com/docs/_mcp

If your documentation has access control enabled, your authenticated endpoint is:

• https://docs.yourdomain.com/_mcp/auth
• https://yourdomain.com/docs/_mcp/auth

Partially private sites expose both endpoints: /_mcp for public content and /_mcp/auth for gated content.

Understand what it can access

The Reader MCP Server serves your published documentation. It sits in front of your published site, not your authoring environment.

For public documentation, the server exposes content that is open to anyone. No authentication is required.

For private, partial, or password-protected documentation, the /_mcp/auth endpoint supports two authentication types:

• OAuth 2.1: For interactive connections where a user is present. The client opens a browser, the user approves the connection and signs in through your site's existing authentication method (password, JWT, or OAuth 2.0), and a credential is minted automatically.
• Machine-to-machine credentials: For automated access without a browser. An admin creates a credential in the dashboard that produces a token with a dmcp_ prefix. The token authenticates directly with the /_mcp/auth endpoint, scoped by the access roles assigned at creation time.

Both types produce a token the client uses for subsequent requests. Access is scoped by the roles assigned to the credential or OAuth session.

Because the server is read-only, it works well when you want grounded answers from trusted documentation while keeping editing controls entirely separate.

Authentication for private documentation

When your documentation has access control enabled, the /_mcp/auth endpoint supports two authentication types: OAuth 2.1 for interactive client connections, and machine-to-machine credentials for automated access without a browser.

OAuth 2.1

OAuth 2.1 works with any authentication method you have configured (password, JWT, or OAuth 2.0). The flow is handled automatically by supported MCP clients.

Built-in support for Claude Desktop, Cursor, VS Code, Claude Web, and ChatGPT connectors means these clients handle the OAuth flow automatically. No manual token management is needed for those clients.

Callback URLs

For MCP clients that use OAuth redirect URIs, manage the allowed callback URLs from the dashboard. Built-in URIs for Claude, ChatGPT, and Cursor are included by default. Add custom URIs for other clients as needed. Up to 50 callback URLs are supported.

Machine-to-machine credentials

For clients that do not support OAuth 2.1 directly, or for automated access where no browser is available, admins can create and manage machine-to-machine credentials from the dashboard. Each credential produces a token with a dmcp_ prefix that authenticates directly with the /_mcp/auth endpoint.

Create a credential

1. Go to MCP Servers in the dashboard sidebar.
2. Under the Reader MCP panel, click Create Credential.
3. Enter a name, select access roles, and choose an expiry: 30 days, 60 days, 90 days, 180 days, 1 year, or never.
4. Click Create to generate the credential.

The token is shown once after creation. Copy it immediately and store it securely. A ready-made client configuration JSON is also provided so you can paste it directly into your MCP client.

View and edit credentials

All credentials appear in a table showing the name, source, access roles, last used time, and expiry. You can edit access roles on an existing credential without re-issuing the token.

Revoke a credential

Revoke individual credentials from the credential table, or revoke all credentials at once. Revocation takes effect immediately.

Configure your MCP client

Add your endpoint to any MCP-compatible client. Replace docs.yourdomain.com with your own documentation domain.

{
  "mcpServers": {
    "my-docs": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://docs.yourdomain.com/_mcp",
        "--timeout",
        "120000"
      ]
    }
  }
}

claude mcp add my-docs \
  --transport http \
  --url https://docs.yourdomain.com/_mcp

{
  "mcpServers": {
    "my-docs": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://docs.yourdomain.com/_mcp",
        "--timeout",
        "120000"
      ]
    }
  }
}

For Claude Desktop, add the configuration to claude_desktop_config.json. For Cursor, add it to your mcp.json. For Claude Code, run the command from your terminal, or add the same JSON to a .mcp.json file in your project root.

Available tools

The Reader MCP Server exposes your published documentation through a retrieval tool, searchDocumentation. Connected clients use it to find the most relevant pages, headings, API reference sections, and snippets in your docs.

Your MCP client discovers the tool and its parameters automatically when it connects, so you do not need to configure anything beyond the endpoint itself.

Best-fit use cases

Choose the Reader MCP Server when the goal is documentation retrieval, not documentation management.

• Customer-facing documentation: Let customers connect AI tools to your published docs for product questions, onboarding, and API lookup.
• Support and success workflows: Help your team answer from the same published docs your customers use.
• AI assistants and copilots: Add a retrieval layer over your docs so responses stay tied to current published content.

Troubleshooting

What to use next