claude-context-local

The claude-context-local MCP server provides a fully local, privacy-first code context assistant that uses a PostgreSQL database with pgvector for vector storage and Ollama for local embeddings. It replaces cloud dependencies with local components to ensure your code and data never leave your machine. The server is configured to connect to a local PostgreSQL database and a locally running Ollama service to generate embeddings from your code and related content. You can run it directly with Node.js using the provided index-postgres.js entry point, or install via npm and run with the included CLI tool. Tools exposed by this MCP include indexing a local codebase, performing semantic search over indexed code, and serving code context suggestions during your coding workflow. The configuration example demonstrates how to wire the MCP into Claude Code or other clients, including embedding provider, Ollama model, and PostgreSQL connection details.

Prerequisites:

• PostgreSQL installed and pgvector extension available
• Ollama installed and running with the required embedding model
• Node.js installed

Installation steps:

1. Clone the repository: git clone https://github.com/MikeO-AI/claude-context-local.git cd claude-context-local

2. Install dependencies (Node.js project): pnpm install

3. Build the MCP server package (PostgreSQL-based): pnpm build

4. Ensure external services are running:

   • PostgreSQL with database named embeddings and pgvector extension
   • Ollama server running and the model DC1LEX/nomic-embed-text-v1.5-multimodal available

5. Run tests (optional): node test-postgres.js

6. Run the MCP server (direct method): node packages/mcp/dist/index-postgres.js

7. Alternative: install the MCP CLI globally and run via CLI: npm install -g @mikeo-ai/claude-context-local-mcp claude-context-local-mcp

8. Optional: run with environment variables for custom configuration, e.g.: POSTGRES_HOST=your-db POSTGRES_PASSWORD=your-pass node packages/mcp/dist/index-postgres.js

Tips:

• Ensure Ollama is accessible at the configured host and port; if you change the host, update OLLAMA_HOST in the environment.
• If pgvector extension is missing, install it and create the extension in your embeddings database.
• The MCP can be configured for Claude Code integration by embedding the server details in your Claude config JSON under mcpServers.
• When using Docker, you can map local PostgreSQL and Ollama services to the containerized MCP as needed. Check MCP-SETUP-GUIDE.md for additional setup guidance.
• Common issues: model not found in Ollama, connection failures to PostgreSQL, or missing vector extension. Validate each service independently before starting the MCP.