docs

The PipeCD Docs MCP Server provides a local MCP-based interface to search and retrieve PipeCD documentation. It clones the official PipeCD docs repository, indexes Markdown files by extracting titles and content, and exposes two MCP methods: search_docs for full-text search across the docs, and read_docs to fetch the content of a specific page. The search is designed to be lightweight and fast, suitable for local development and offline-like workflows. To use it, run the server via MCP (for example with npx as described in the installation notes) and then issue MCP calls from your MCP client to query documents or fetch page content. The indexing happens in a temporary directory and is kept in sync as needed by the server implementation.

Available tools:

• search_docs: Performs a full-text search over the docs with parameters such as query, offset, and limit.
• read_docs: Returns the content of a specified document by its path relative to the docs content folder. This is useful for rendering pages in an app or extracting metadata for display.

This setup is intended for local development and integration with MCP-enabled tooling. It provides a simple, focused API surface: search by keywords and retrieve documents by path, making it easy to build documentation search UIs or developer tooling around PipeCD docs.

Prerequisites:

• Node.js (recommended: 14.x or newer) and npm/yarn installed on your system.
• A Personal Access Token (PAT) with read:packages scope if you plan to access private npm packages or use a private registry as described in the docs.
• Optional: Cursor for quick one-click installation through the provided deeplink (see README).

Installation steps:

1. Prepare npm registry authentication (if required):

@pipe-cd:registry=https://npm.pkg.github.com
//npm.pkg.github.com/:_authToken=YOUR_TOKEN

2. Install or run via npx (recommended for MCP servers):

# Run the MCP server directly using npx
npx @pipe-cd/docs-mcp-server@latest

3. If integrating with Cursor or other installers, follow the specific installation flow provided (see README’s Cursor deeplink guidance).

4. If you’re using an MCP config file (mcp.json), add the server entry as shown in the example:

{
  "mcpServers": {
    "pipe-cd.docs-mcp-server": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "@pipe-cd/docs-mcp-server@latest"
      ]
    }
  }
}

Prerequisites recap:

• Node.js and npm/yarn installed
• Access tokens configured if needed for private npm registries
• MCP client capable of sending MCP protocol requests to the server

Tips and common issues:

• If you encounter slow cloning, the server uses sparse-checkout to minimize clone size. Ensure git is configured to allow sparse-checkout for optimal performance.
• The server indexes Markdown files by their titles and content. If titles aren’t present in front matter, consider updating docs or contributing to the server to adjust the extraction logic.
• Ensure your npm registry authentication is valid; invalid tokens will prevent installing the MCP server package via npx.
• For local debugging, you can run the MCP server and test with a basic MCP client to verify search_docs and read_docs behavior before integrating into larger tooling.
• If you need to customize where docs are cloned or how indexing behaves, check the implementation notes and environment configuration in the server’s codebase.