MCP Snippets Server

An MCP (Model Context Protocol) server that provides semantic search capabilities for code snippets using RAG (Retrieval-Augmented Generation) and vector embeddings.

Overview

This server processes Markdown files containing code snippets, creates vector embeddings for semantic search, and provides an MCP tool for finding relevant code snippets based on topic queries.

Features

• Vector Store: Creates and manages a persistent vector store from Markdown documentation
• Semantic Search: Uses OpenAI-compatible embeddings to find relevant snippets
• MCP Integration: Exposes search functionality as an MCP tool
• Automatic Processing: Processes .md files on first run and stores embeddings
• Persistent Storage: Saves vector store to JSON for quick subsequent startups

Architecture

The server consists of several key components:

• Vector Store (rag/): Handles embedding storage and similarity search
• Helpers (helpers/): File processing utilities
• Snippets (snippets/): Contains code snippet documentation in Markdown format
• MCP Server: Exposes the search_snippet tool via HTTP

Configuration

Set the following environment variables (or use .env file):

• MODEL_RUNNER_BASE_URL: OpenAI-compatible API endpoint (default: http://localhost:12434/engines/llama.cpp/v1/)
• EMBEDDING_MODEL: Embedding model name (default: ai/mxbai-embed-large:latest)
• JSON_STORE_FILE_PATH: Vector store file path (default: rag-memory-store.json)
• MCP_HTTP_PORT: HTTP server port (default: 9090)
• LIMIT: Similarity threshold (default: 0.6)
• MAX_RESULTS: Maximum search results (default: 2)

Usage

Starting the Server

go run main.go

The server will:

1. Load existing vector store or create new one from .md files
2. Start HTTP server on the configured port
3. Expose MCP endpoint at /mcp

MCP Tool

The server provides one MCP tool:

• search_snippet: Find code snippets related to a topic
  • Parameter: topic (string) - Search query or question

Example Tool Call

{
  "method": "tools/call",
  "params": {
    "name": "search_snippet",
    "arguments": {
      "topic": "how to create a REST API in Go"
    }
  }
}

Development

File Structure

• main.go: Main server implementation
• rag/: Vector store and similarity search logic
• helpers/: File processing utilities
• snippets/: Code snippet documentation
• store/: Persistent vector store data

Adding New Snippets

1. Add Markdown files to the snippets/ directory
2. Use ---------- as delimiter between different snippets
3. Restart the server to reprocess and update embeddings

Dependencies

• github.com/mark3labs/mcp-go: MCP server implementation
• github.com/openai/openai-go/v2: OpenAI API client for embeddings
• github.com/joho/godotenv: Environment variable management
• github.com/google/uuid: UUID generation

Docker Support

The project includes Docker configuration for easy deployment.

Use the Docker Image

Image: https://hub.docker.com/repository/docker/k33g/mcp-snippets-server/tags

1. In a directory, create a snippets folder and add your .md files with code snippets.
2. Create a compose.yml file with the following content:

services:

  mcp-snippets-server:
    image: k33g/mcp-snippets-server:0.0.5
    ports:
      - 9090:6060
    environment:
      - MCP_HTTP_PORT=6060
      - LIMIT=0.6
      - MAX_RESULTS=2
      - JSON_STORE_FILE_PATH=store/rag-memory-store.json
      - DELIMITER=----------
    volumes:
      - ./snippets:/app/snippets
      - ./store:/app/store

    models:
      mxbai-embed:
        endpoint_var: MODEL_RUNNER_BASE_URL
        model_var: EMBEDDING_MODEL

models:
  mxbai-embed:
    model: ai/mxbai-embed-large:latest

Start the server with:

docker compose up