MCP Memory Service

# MCP Memory Service — Local Setup and Run Follow these steps to run the service locally, switch storage backends, and validate functionality. ## 1) Install Dependencies Using uv (recommended): ``` uv sync ``` Using pip: ``` python -m venv .venv source .venv/bin/activate # Windows: .venv\Scripts\activate pip install -e . ``` If using SQLite-vec backend (recommended): ``` uv add sqlite-vec sentence-transformers torch # or pip install sqlite-vec sentence-transformers torch ``` ## 2) Choose Storage Backend SQLite-vec (default): ``` export MCP_MEMORY_STORAGE_BACKEND=sqlite_vec # optional custom DB path export MCP_MEMORY_SQLITE_PATH="$HOME/.local/share/mcp-memory/sqlite_vec.db" ``` ChromaDB (deprecated): ``` export MCP_MEMORY_STORAGE_BACKEND=chroma export MCP_MEMORY_CHROMA_PATH="$HOME/.local/share/mcp-memory/chroma_db" ``` Cloudflare: ``` export MCP_MEMORY_STORAGE_BACKEND=cloudflare export CLOUDFLARE_API_TOKEN=... export CLOUDFLARE_ACCOUNT_ID=... export CLOUDFLARE_VECTORIZE_INDEX=... export CLOUDFLARE_D1_DATABASE_ID=... ``` ## 3) Run the Server Stdio MCP server (integrates with Claude Desktop): ``` uv run memory server ``` FastMCP HTTP server (for Claude Code / remote): ``` uv run mcp-memory-server ``` Configure Claude Desktop example (~/.claude/config.json): ``` { "mcpServers": { "memory": { "command": "uv", "args": ["--directory", "/path/to/mcp-memory-service", "run", "memory", "server"], "env": { "MCP_MEMORY_STORAGE_BACKEND": "sqlite_vec" } } } } ``` ## 4) Verify Health and Basic Ops CLI status: ``` uv run memory status ``` MCP tool flow (via client): - store_memory → retrieve_memory → search_by_tag → delete_memory ## 5) Run Tests ``` pytest -q # or uv run pytest -q ``` See also: `docs/mastery/testing-guide.md` and `docs/sqlite-vec-backend.md`.