GDS Agent MCP
A Neo4j MCP server and skill: the agent runs Graph Data Science algorithms on a graph, from centrality to embeddings
- 99 ★ on the repository
- Last commit
Extended access
Why this level
- The server connects to your Neo4j database with a login and password and reads its data
- Database writes open only with `NEO4J_READ_ONLY=false`, so keep the default
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/gds-agentDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server connects to your Neo4j database with a login and password and reads its data
- Database writes open only with `NEO4J_READ_ONLY=false`, so keep the default
This is third-party code. Review the repository files before installing.
What it does
GDS Agent has two parts. The MCP server exposes Neo4j Graph Data Science algorithms: centrality, community detection, path finding, similarity, node embeddings and machine learning pipelines. The `neo4j-graph-data-scientist` skill teaches the agent which algorithm to pick and how to run graph analysis by good practice. The server works with the GDS plugin on a self-managed Neo4j or with Aura Graph Analytics, where the agent creates a session itself with `create_session` and `delete_session`. For free queries there are `read_neo4j_cypher` and `get_neo4j_schema`, while `write_neo4j_cypher` stays hidden until `NEO4J_READ_ONLY` is `false`. Output is capped at 500 rows and 100 thousand characters to keep the context small.
Who it is for. Data analysts and engineers whose data sits in Neo4j and who prefer asking graph questions through an agent.
Good fit when
- Influential nodes and communities must be found in a graph without hand-writing GDS procedures
- A shortest path between nodes is needed in a station or link graph
- Node embeddings are needed and the agent must choose the algorithm and parameters
- The graph schema must be studied before questions, and `get_neo4j_schema` fits
Not a fit when
- There is no data in Neo4j: the server does not run without a database
- Writing to the database without confirmation is needed: writes are hidden until read-only mode is lifted
- Another graph engine is needed: the server only knows Neo4j
Example request
Find the five most central stations in the metro graph by PageRank and the shortest path between Victoria and King's CrossMCP
- Transport
- stdio, http
- Authentication
- API key
Environment variables
NEO4J_URIrequired- Connection URI `neo4j://` or `neo4j+s://`.
NEO4J_USERNAMErequired- Database user name.
NEO4J_PASSWORDrequired, secret- Database user password.
NEO4J_READ_ONLY- When `false` opens the Cypher write tool; writes are hidden by default.
Limitations
`NEO4J_URI`, `NEO4J_USERNAME` and `NEO4J_PASSWORD` are required to reach Neo4j. The algorithms need the GDS plugin on your own Neo4j server or Aura Graph Analytics with Aura API credentials. `get_neo4j_schema` uses APOC. The server launches through `uvx gds-agent` and needs uv installed. Database writes are available only after `NEO4J_READ_ONLY=false`. An Aura session gets 8 GB of memory and lives 12 hours by default, which costs Aura resources.
How to disable. In Claude Code run `/plugin uninstall gds-agent@neo4j-gds`; in other clients remove the `neo4j-gds` entry from the MCP config.
FAQ
How do I install GDS Agent in Claude Code?
With `/plugin marketplace add neo4j-contrib/gds-agent` and `/plugin install gds-agent@neo4j-gds`. The skill installs with the plugin.
How do I connect it in Codex?
`codex mcp add neo4j-gds -- uvx gds-agent` for the server and `npx skills add neo4j-contrib/gds-agent -a codex` for the skill.
What does the database need for algorithms?
The GDS plugin on your own Neo4j or Aura Graph Analytics. The server detects the mode at startup through a `gds.session.list()` call.
Can the agent change data?
Not by default: `write_neo4j_cypher` is hidden until `NEO4J_READ_ONLY` is `false`.
What license applies?
MIT.
Discussion
No questions or reviews yet.
Related
MCP serversNeo4j Labs MCP servers: Cypher queries on a graph, graph memory for agents, Aura management and data modeling
Neo4j skills for agents: Cypher queries, graph modeling, drivers, import, GraphRAG, GDS and Aura provisioning
A knowledge graph with an MCP server: the agent searches concepts by meaning and sees how well each is supported.
GraphJin MCP
GraphJin
GraphJin's MCP server gives an agent safe, schema-aware access to SQL and MongoDB instead of raw database credentials
An agent queries GreptimeDB metrics, logs and traces with SQL, TQL and RANGE, with write protection and data masking
An MCP server for PostgreSQL with several databases at once, table descriptions from the catalog, PostGIS and pgvector hints and query plans
README in short
The two-part structure, an install table by client, an environment variable reference, HTTP mode and Aura sessions. Then Cypher tools, a London metro example dataset, running for development and contribution rules.