RAG Documentation

You run an MCP server that exposes tools to index, search, and manage documentation via vector search. It lets you add sources, index local repositories, monitor progress, and use a web interface to test queries and monitor queues in real time.

How to use

Use an MCP client to interact with the rag-docs server to search documentation, manage sources, and index repositories. You can perform direct actions on individual documentation sources or batch process multiple sources through a queue. Monitor indexing progress and status via the web interface or the MCP tools you enable in your client.

How to install

Prerequisites: you need Node.js installed on your machine and a working environment for running MCP servers. You will also set up environment variables for embedding models and an optional cloud fallback.

Follow these concrete steps to install and run the rag-docs MCP server locally.

1. Install dependencies in a project that will host the MCP server. Depending on your setup, you may clone a repository or prepare a working directory. Then install required packages.

Configuration and usage notes

This MCP server uses a stdio-based local server entry that runs via Node and points to the built index file. You configure embedding providers, models, and a queue of tools you want to auto-approve for use.

Detailed start and configuration example

{
  "mcpServers": {
    "rag-docs": {
      "type": "stdio",
      "name": "rag-docs",
      "command": "node",
      "args": ["/path/to/your/mcp-ragdocs/build/index.js"],
      "env": {
        "EMBEDDING_PROVIDER": "ollama",
        "EMBEDDING_MODEL": "nomic-embed-text",
        "OPENAI_API_KEY": "your-api-key-here",
        "FALLBACK_PROVIDER": "openai",
        "FALLBACK_MODEL": "nomic-embed-text",
        "QDRANT_URL": "http://localhost:6333"
      },
      "disabled": false,
      "autoApprove": [
        "search_documentation",
        "list_sources",
        "extract_urls",
        "remove_documentation",
        "list_queue",
        "run_queue",
        "clear_queue",
        "add_documentation",
        "add_repository",
        "list_repositories",
        "update_repository",
        "remove_repository",
        "watch_repository",
        "get_indexing_status"
      ]
    }
  }
}

Security and operation notes

Maintain proper access to the web interface and ensure your OpenAI API key and any cloud keys are stored securely. Use the queueing features to batch large documentation ingestions and avoid long-running requests in a single MCP interaction.

Troubleshooting tips

If the web interface cannot load or the queue shows stalled items, check that the MCP process is running and that the embedding provider services are reachable. Review environment variables and ensure the correct file paths exist for the built index script.

Repository and indexing notes

The server supports indexing local code repositories with configurable include/exclude patterns and per-file chunking strategies. You can watch repositories for changes to automatically re-index updated files and remove deleted files from the index.

Examples

Example usage to register a local repository by name with watch mode enabled.

Appendix: Tools and capabilities

This server provides a suite of tools that enable you to search documentation, list sources, extract URLs, manage queues, and index both web content and local repositories. See the Tools section for a complete list and descriptions.

Available tools

search_documentation

Search through the documentation using vector search and return relevant chunks with source information

list_sources

List all available documentation sources with metadata

extract_urls

Extract URLs from text and check for duplicates in the documentation set

remove_documentation

Remove documentation from a specific source to clean up outdated content

list_queue

List all items in the processing queue with status information

run_queue

Process all items in the queue and index new content into the vector store

clear_queue

Clear all items from the processing queue to reset processing state

add_documentation

Add new documentation directly from a URL; fetches, processes, and indexes content immediately

add_repository

Index a local repository with include/exclude patterns and per-file chunking; uses asynchronous processing and provides heartbeat logging

list_repositories

List all indexed repositories with their configurations and watch status

update_repository

Re-index a repository with updated settings and configurations

remove_repository

Remove a repository from the index and delete associated documents

watch_repository

Start or stop watching a repository for changes and auto-update the index

get_indexing_status

Get current indexing progress and status for one or all repositories