Mimir

MCP Integration

Connect AI assistants to Mimir using Model Context Protocol

What is MCP?

Model Context Protocol (MCP) is an open protocol that enables AI assistants to securely access external data sources and tools. Mimir provides an MCP server that connects your AI coding assistant directly to your documentation and codebase.

Features

  • Configurable Backend: Set your own Mimir RAG backend URL via environment variables
  • Customizable Server Name: Configure the MCP server name to match your organization
  • Fast Semantic Search: Direct access to your documentation
  • No Additional LLM Calls: Returns document chunks directly
  • Works with Popular Assistants: Claude Code, VS Code, Claude Desktop, and more

Important Notes

Embeddings: The Mimir backend uses its own embedding model. You cannot use a different LLM provider for embeddings as this would cause compatibility issues.

Chat Completions: When using chat completions via the API, you'll need to provide your own LLM API keys (OpenAI, Anthropic, etc.) for the completion model. The embedding model is handled by the backend.

Installation

VS Code / Claude Code

  1. Open Command Palette (Cmd+Shift+P or Ctrl+Shift+P)
  2. Select "MCP: Open User Configuration"
  3. Add:
{
  "mcpServers": {
    "mimir": {
      "command": "npx",
      "args": ["-y", "@your-org/your-mcp-name"],
      "env": {
        "MIMIR_API_URL": "https://your-backend-url.com",
        "MCP_SERVER_NAME": "Your Custom MCP Server Name"
      }
    }
  }
}

Note: Replace @your-org/your-mcp-name with your published npm package name, and set your backend URL and server name.

  1. Restart VS Code or reload the MCP extension

Claude Desktop

Edit claude_desktop_config.json:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "mimir": {
      "command": "npx",
      "args": ["-y", "@your-org/your-mcp-name"],
      "env": {
        "MIMIR_API_URL": "https://your-backend-url.com",
        "MCP_SERVER_NAME": "Your Custom MCP Server Name"
      }
    }
  }
}

Note: Replace @your-org/your-mcp-name with your published npm package name, and set your backend URL and server name.

Local Development

For local development:

  1. Clone and Build:
cd your-mcp-directory
npm install
npm run build
  1. Set Backend URL: Set the MIMIR_API_URL environment variable:
export MIMIR_API_URL=http://localhost:3000
export MCP_SERVER_NAME="My Local MCP Server"
  1. Add to MCP Config:
{
  "mcpServers": {
    "mimir": {
      "command": "node",
      "args": ["/absolute/path/to/your-mcp/dist/index.js"],
      "env": {
        "MIMIR_API_URL": "http://localhost:3000",
        "MCP_SERVER_NAME": "My Local MCP Server"
      }
    }
  }
}

Using the askDocs Tool

Once configured, your AI assistant has access to the askDocs tool. Simply ask questions:

  • "Find documentation about authentication"
  • "Search the docs for API endpoint examples"
  • "What does the documentation say about error handling?"

The AI assistant will automatically:

  1. Invoke the askDocs tool
  2. Retrieve relevant documentation chunks
  3. Synthesize an answer based on the retrieved content

Tool Parameters

The askDocs tool accepts these parameters:

  • question (required): Your documentation search query
  • matchCount (optional): Number of chunks to retrieve (default: 10)
  • similarityThreshold (optional): Minimum similarity score (default: 0.2)

How It Works

  1. AI Assistant receives your question
  2. MCP Server (your custom MCP package) receives the tool invocation
  3. Backend (mimir-rag) performs semantic search using embeddings
  4. Document Chunks are returned with content and metadata
  5. AI Assistant synthesizes an answer from the retrieved content

Prerequisites

  • A running mimir-rag server with ingested documentation
  • Node.js 20 or later
  • An AI assistant that supports MCP

Troubleshooting

Connection Issues

If the MCP server can't connect:

  1. Verify mimir-rag is running: curl https://your-backend-url.com/health
  2. For local development, ensure MIMIR_API_URL is set to http://localhost:3000
  3. Check firewall and network settings
  4. Verify the MIMIR_API_URL or MCP_BACKEND_URL environment variable is correctly set in your MCP configuration

Tool Not Available

If your AI assistant doesn't see the tool:

  1. Restart your AI assistant
  2. Verify MCP configuration JSON is valid
  3. Check MCP logs for errors
  4. Ensure the path to dist/index.js is absolute (for local dev)

Empty Results

If searches return no results:

  1. Verify documentation has been ingested
  2. Check embeddings were created successfully
  3. Try lowering similarityThreshold for broader matches
PreviousAPI Reference