VeritasGraph MCP
A local GraphRAG MCP server: it builds a knowledge graph from documents and answers with source citations through Ollama.
- 327 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- In the default `write` mode the agent changes the knowledge graph and `admin` clears it.
- Ingested document content goes into the model context.
Install
- Works in
Manual install
pip install veritasgraph-mcpThe PyPI package, run with `python -m veritasgraph_mcp`. A local Ollama with a model is required.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- In the default `write` mode the agent changes the knowledge graph and `admin` clears it.
- Ingested document content goes into the model context.
This is third-party code. Review the repository files before installing.
What it does
VeritasGraph is an open-source GraphRAG engine, and its `veritasgraph-mcp` MCP server connects the local engine to Claude Desktop, Cursor, VS Code, Windsurf, Cline and Continue over JSON-RPC on stdio. It runs locally with no cloud calls or telemetry, and the only network hop goes to the local Ollama runtime. Run it with `python -m veritasgraph_mcp`. Tools are tiered by the `VERITASGRAPH_MCP_MODE` variable: `read`, `write` (the default) and `admin`. Read: `veritasgraph_query` gives a multi-hop answer with citations, a reasoning path and an `answer_id`, plus `veritasgraph_search_entities`, `veritasgraph_get_entity`, `veritasgraph_get_graph`, the ontology and glossary terms. Write: `veritasgraph_ingest_document` chunks a document and merges entities into the graph, `veritasgraph_import_graph` and `veritasgraph_propose_term`. Admin: `veritasgraph_clear_graph` clears the graph after taking a snapshot.
Who it is for. Teams that need a knowledge graph over their own documents with verifiable citations and no data leaving the machine.
Good fit when
- Company documents are loaded into a graph and multi-hop questions are asked with source citations.
- Before spending a model call it is checked whether the graph can answer the question.
- One entity is examined with its relations and source text chunks.
- A ready `graph.json` from another tool is imported into the graph.
Not a fit when
- There is no local Ollama and model: ingest and answers need them.
- The agent must not change the graph: keep `read` mode, since writing is on by default.
- A cloud service with ready infrastructure is needed: the engine is deployed on your own machine.
Example request
Ingest this document into the graph, then answer with `[doc#chunk]` citations which data storage requirements it describes.MCP
- Transport
- stdio
- Authentication
- not required
Environment variables
VERITASGRAPH_MCP_MODE- Tool tier: read, write (default) or admin
VERITASGRAPH_MODEL- Default Ollama model, for example qwen3:latest
OLLAMA_HOST- Local Ollama address, 127.0.0.1:11434 by default
Limitations
Python 3.10 or newer, the VeritasGraph repository dependencies and a local Ollama with a pulled model such as `qwen3:latest` are required: ingest and answers go through it, while entity search and graph reads do not call the model. The default mode is `write`, so write tools are available at once, and `admin` enables graph clearing with a snapshot. Tools outside the active tier are hidden and return a `-32601` error. The governed SQL search layer needs the `structured` extra package.
How to disable. Remove the `veritasgraph` entry from the client. The graph lives in the project shared SQLite store and is cleared separately, in `admin` mode with `veritasgraph_clear_graph`.
FAQ
How do I run the server?
From the repository root: `python -m veritasgraph_mcp`, or `python -m veritasgraph_mcp --debug` for verbose logging. The PyPI package is `veritasgraph-mcp`.
What are the tiers?
The `VERITASGRAPH_MCP_MODE` variable selects `read`, `write` (the default) or `admin`, and each tier includes the one below.
What is needed for answers?
A local Ollama and a pulled model, for example `ollama pull qwen3:latest`. `VERITASGRAPH_MODEL` sets the default model.
Does data go to the cloud?
No, per the project: there are no cloud calls or telemetry and the only network hop goes to the local Ollama.
What does a query return?
An answer with `[doc#chunk]` citations, a reasoning path and an `answer_id` that `veritasgraph_get_answer_receipt` later replays.
Discussion
No questions or reviews yet.
Related
MCP serversGraphiti MCP
Graphiti MCP Server
Zep's Graphiti MCP server: long-term agent memory as a time-aware knowledge graph on FalkorDB or Neo4j
An MCP server that splits math papers from arXiv, LaTeX and PDF into theorems, proofs and dependencies backed by the source
A CLI and MCP server that builds a local wiki with a knowledge graph from documents, code and links, so the agent searches it instead of scanning files
A personal LLM wiki with a knowledge graph, cited RAG and memory, served to Claude Code and Cursor by an MCP server
An MCP server that keeps domain knowledge in a graph: the agent searches entities and relations and can add or update them itself.
MCP Knowledge Graph
mcp-knowledge-graph
Persistent memory for AI in a local knowledge graph: entities, relations and observations are stored in .aim files on your disk
README in short
The server README covers tools by tier, `veritasgraph://` resources, requirements, running and environment variables. The project README is about the GraphRAG engine and the Studio interface.