mcp-snippets

This MCP server processes Markdown files containing code snippets, builds a vector store from them, and exposes an MCP tool named search_snippet. It uses semantic embeddings to perform retrieval-augmented search over the code snippets and returns the most relevant matches. To use it, start the server and then call the MCP tool via the provided /mcp endpoint. The available tool is search_snippet, which accepts a topic string and returns code snippets that semantically relate to that topic.

Prerequisites:

• Go installed (go1.20+ recommended)
• Access to a starting directory containing a snippets/ folder and Markdown files

Installation steps:

1. Clone the repository or download the source code.
2. Ensure your environment variables are set (see mcp_config below) or create a .env file.
3. Build/run the server:

# From the project root
go run main.go

If you prefer Docker, see the Docker instructions in the README and use the provided docker-compose setup with the mcp-snippets-server image. The server will process Markdown files on first run and persist embeddings to the configured JSON store.

Tips and common considerations:

• The vector store is persisted to JSON (default rag-memory-store.json) to speed up subsequent startups.
• Update the snippets/ directory by adding or removing Markdown files and restart the server to reprocess and refresh embeddings.
• Configure OPENAI-compatible embeddings via EMBEDDING_MODEL and MODEL_RUNNER_BASE_URL for local or cloud embeddings.
• If using Docker, ensure the MCP_HTTP_PORT mapping matches your environment and consider using the provided compose.yml for quick deployment.
• The MCP tool endpoint is exposed at /mcp, and you can call search_snippet with a topic string to retrieve relevant code snippets.