Cursor-history

Cursor-history MCP provides a Dockerized FastAPI server that exposes a simple API to search your vectorized Cursor chat history. After running the extraction script to generate a LanceDB database of prompts and embeddings, you start the API in a container. The server exposes endpoints such as POST /search_chat_history for vector similarity search and GET /health to verify connectivity to Ollama and LanceDB. You can use these endpoints to perform Retrieval Augmented Generation (RAG) tasks by passing a query and receiving matching historical prompts with associated metadata. The workflow is designed to keep computation local: embeddings are generated using a local Ollama model (nomic-embed-text) and stored in LanceDB, with the API serving search results from that database.

Prerequisites:\n- Docker Desktop (Windows/Mac) or Docker Engine (Linux) installed and running.\n- Python 3.7+ (for extraction script, if you plan to run the extractor locally).\n- Ollama installed and running with nomic-embed-text:latest model available.\n\nInstall and set up steps:\n1) Clone the repository and navigate to the project root.\n2) If you plan to run the extraction locally:\n - Install Python dependencies for the extractor:\n pip install -r requirements.txt\n - Ensure WORKSPACE_STORAGE_PATH in cursor_history_extractor.py points to your Cursor data.\n - Start Ollama and verify nomic-embed-text:latest is available: ollama pull nomic-embed-text:latest; ollama run nomic-embed-text:latest.\n3) Run the extraction script to create the LanceDB database (host machine):\n python cursor_history_extractor.py\n This creates ./cursor_chat_history.lancedb on the host.\n4) Build and run the API Docker container (as shown in the MCP config):\n docker build -t cursor-chat-search-api .\n docker run -i -p 8001:8001 -v /path/to/your/cursor_chat_history.lancedb:/data/cursor_chat_history.lancedb -e OLLAMA_HOST="http://host.docker.internal:11434" cursor-chat-search-api\n\nNotes:\n- The volume mount must point to the absolute path of the LanceDB directory on the host so the container can access the database.\n- If running on Linux, adjust OLLAMA_HOST and networking as needed to allow the container to reach the Ollama instance.\n- Ensure the LanceDB table used is named chat_history and that the database path inside the container matches /data/cursor_chat_history.lancedb.

Tips and common issues:\n- Ensure Ollama is reachable from inside the Docker container. If you run Ollama on the host, use the recommended host networking or the host.docker.internal alias (Windows/Mac).\n- The extraction script writes to ./cursor_chat_history.lancedb by default; make sure you have write permissions in the target directory.\n- The API exposes /health to check connections to Ollama and LanceDB; use it to confirm that the database is open and embeddings can be retrieved.\n- When mounting volumes on Windows, use forward slashes and absolute paths, e.g. C:/path/to/cursor_chat_history.lancedb:/data/cursor_chat_history.lancedb.\n- If you change the LanceDB directory, update the mapping in the Docker run command accordingly.\n- The embedding model nomic-embed-text:latest defaults to 768 dimensions; ensure the model version in Ollama matches what the extractor expects.