A model that knows a property is a string will guess a value, and a guess that misses returns nothing — which reads like an empty result rather than a mistake. Sampling the actual distinct values and shipping them in the schema removes that whole failure mode. High-cardinality fields, IDs, UUIDs, timestamps and long text are excluded automatically, and GREMLIN_SCHEMA_INCLUDE_SAMPLE_VALUES stays off by default to keep the schema small.
A graph database server for Apache TinkerPop and anything else that speaks Gremlin. It discovers the schema, executes traversals, and imports and exports data.
- `get_graph_schema` returns the full structure — node types, edge types and how they connect
- `run_gremlin_query` executes any Gremlin traversal with full syntax support
- `get_graph_status` verifies connectivity and server status; `refresh_schema_cache` forces the cached schema to reload
- `import_graph_data` loads GraphSON, CSV or JSON with batch processing; `export_subgraph` extracts to JSON, GraphSON or CSV
- Automatic enum discovery scans low-cardinality properties and puts their real values in the schema, so a status filter uses CONFIRMED rather than a guessed 'active'
The npm package @kpritam/gremlin-mcp, run with npx, on Node.js 18.0.0 or newer. GREMLIN_ENDPOINT points at your server, for example localhost:8182; GREMLIN_USERNAME, GREMLIN_PASSWORD and GREMLIN_USE_SSL cover authenticated deployments. Enum discovery is tunable: GREMLIN_ENUM_CARDINALITY_THRESHOLD defaults to 10, GREMLIN_ENUM_PROPERTY_BLACKLIST excludes properties, and GREMLIN_ENUM_DISCOVERY_ENABLED turns it off.
One command — npx @kpritam/gremlin-mcp