Docs-to-MCP

You can turn an AI Search-based knowledge base into a practical MCP server that responds to client requests. This enables you to query docs chunks, receive AI-generated responses with sources, and connect from common MCP clients using a simple, remote HTTP endpoint or a local CLI driver.

How to use

You connect to the MCP server from your MCP client or runner. Use the remote HTTP endpoint for standard MCP transport, or run a local driver that connects to the same MCP URL to bring your own client into the MCP ecosystem.

How to install

Prerequisites: ensure you have Node.js and npm installed on your machine. You will also need pnpm if you follow the deployment flow exactly as shown here.

# 1) Install dependencies (if you are cloning or setting up locally)
npm install

# 2) Run a local development server (optional, if a dev script is provided)
npm run dev

# 3) Deploy to run the MCP server behind a remote URL (if you follow the provided deployment pattern)
pnpm run deploy

Additional notes

Connect to the MCP server using either the HTTP URL or the local STDIO command shown below. Choose the method that matches your client setup. The HTTP URL is the remote MCP endpoint you expose, while the STDIO approach is for local tooling that consumes the MCP stream directly.

Available tools

search_docs

Search the documentation and return relevant chunks. Parameters: query (required), max_results (optional), rewrite_query (optional), score_threshold (optional)

ai_search_docs

Search documentation and get an AI-generated response with sources. Parameters: Same as search_docs