Skip to main content
Grounded Docs implements the Model Context Protocol (MCP), enabling AI assistants like Claude Desktop to query and index documentation directly from your MCP server.

​
What is MCP?

The Model Context Protocol provides a standardized way for AI applications to access external data sources and tools. Grounded Docs exposes its documentation search and indexing capabilities through MCP, allowing AI assistants to:
  • Search indexed documentation in real-time
  • Index new library versions
  • Manage scraping jobs
  • Fetch and convert web pages to Markdown

​
Architecture

​
MCP Server Instance

The MCP server is created in src/mcp/mcpServer.ts:18 and exposes:
  • Tools: Callable functions that perform operations
  • Resources: URI-based data endpoints for querying state

​
Tool Registration

Tools are registered with schemas defined using Zod validation: 
server.tool(
  "search_docs",
  "Search up-to-date documentation for a library or package",
  {
    library: z.string().trim().describe("Library name."),
    version: z.string().trim().optional().describe("Library version (exact or X-Range, optional)."),
    query: z.string().trim().describe("Documentation search query."),
    limit: z.number().optional().default(5).describe("Maximum number of results."),
  },
  async ({ library, version, query, limit }) => {
    // Tool implementation
  }
);

​
Read-Only Mode

When configured with readOnly: true, Grounded Docs restricts access to mutating operations: Available in Read-Only Mode:
  • search_docs
  • list_libraries
  • find_version
  • fetch_url
Disabled in Read-Only Mode:
  • scrape_docs
  • refresh_version
  • remove_docs
  • list_jobs
  • get_job_info
  • cancel_job
See src/mcp/mcpServer.ts:22 for the read-only configuration logic.

​
Response Format

All MCP tools return structured responses:

​
Success Response

{
  "content": [
    {
      "type": "text",
      "text": "Result content"
    }
  ]
}

​
Error Response

{
  "isError": true,
  "content": [
    {
      "type": "text",
      "text": "Error: Description of what went wrong"
    }
  ]
}

​
Available Tools

Grounded Docs provides the following MCP tools:

​
Read Operations

​
Write Operations

​
Job Management

​
Resources

MCP resources provide URI-based access to server state:

​
docs://libraries

Returns a list of all indexed libraries.

​
docs://libraries/{library}/versions

Returns all indexed versions for a specific library.

​
docs://jobs?status={status} (Write mode only)

Returns all pipeline jobs, optionally filtered by status.

​
docs://jobs/{jobId} (Write mode only)

Returns details for a specific job by ID.

​
Error Handling

Tools implement consistent error handling:
  1. Validation Errors: Thrown when input parameters are invalid
  2. Tool Errors: Thrown when operations fail (library not found, job not found, etc.)
  3. System Errors: Caught and wrapped with user-friendly messages
See src/tools/errors.ts for error type definitions.

​
Telemetry

All MCP tool calls are tracked for analytics (when telemetry is enabled): 
telemetry.track(TelemetryEvent.TOOL_USED, {
  tool: "search_docs",
  context: "mcp_server",
  library,
  version,
  query: query.substring(0, 100), // Truncated for privacy
  limit,
});
See src/mcp/mcpServer.ts:197 for telemetry implementation.

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us