mcp-graph-memory

An MCP server that builds a semantic graph memory from a project directory. It indexes markdown documentation and TypeScript/JavaScript source code into graph structures, then exposes them as MCP tools that any AI assistant can use to navigate and search the project.

Create graph-memory.yaml — paths must be relative to the container filesystem:

server:
  host: "0.0.0.0"
  port: 3000
  modelsDir: "/data/models"

projects:
  my-app:
    projectDir: "/data/projects/my-app"
    docsPattern: "docs/**/*.md"
    codePattern: "src/**/*.{ts,tsx}"
    excludePattern: "node_modules/**,dist/**"

docker run -d \
  --name graph-memory \
  -p 3000:3000 \
  -v $(pwd)/graph-memory.yaml:/data/config/graph-memory.yaml:ro \
  -v /path/to/my-app:/data/projects/my-app:ro \
  -v graph-memory-models:/data/models \
  ghcr.io/prih/mcp-graph-memory

Three mounts: | Mount | Container path | Description | |-------|---------------|-------------| | Config | /data/config/graph-memory.yaml | Your config file (read-only) | | Projects | /data/projects/ | Project directories to index (read-only, unless you use knowledge/tasks/skills — then remove :ro) | | Models | /data/models/ | Embedding model cache — use a named volume so models persist across container restarts |

The embedding model (Xenova/all-MiniLM-L6-v2, ~90MB) is downloaded on first startup. Subsequent starts use the cached model from the volume.

# docker-compose.yaml
services:
  graph-memory:
    image: ghcr.io/prih/mcp-graph-memory
    ports:
      - "3000:3000"
    volumes:
      - ./graph-memory.yaml:/data/config/graph-memory.yaml:ro
      - /path/to/my-app:/data/projects/my-app
      - models:/data/models
    restart: unless-stopped

volumes:
  models:

docker compose up -d

• Web UI: http://localhost:3000