Flexible GraphRAG MCP
MCP server for the Flexible GraphRAG platform: ingest from 14 data sources, hybrid search and AI answers over a knowledge graph
- 197 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The ingest tools take logins and passwords for stores such as Alfresco, S3 and SharePoint straight in their parameters.
- Without a user sign-in, search and answers draw on every document in the store, private ones included.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/flexible-graphrag-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The ingest tools take logins and passwords for stores such as Alfresco, S3 and SharePoint straight in their parameters.
- Without a user sign-in, search and answers draw on every document in the store, private ones included.
This is third-party code. Review the repository files before installing.
What it does
Flexible GraphRAG MCP is an MCP server that connects Claude Desktop and other MCP clients to the Flexible GraphRAG platform. The platform parses documents and builds a knowledge graph and a vector index, while the server hands its capabilities to the agent as tools. The `ingest_documents` tool loads files from 14 sources: local disk, Alfresco, Nuxeo, SharePoint, OneDrive, Box, Google Drive, S3, Google Cloud Storage, Azure Blob, CMIS, web pages, Wikipedia and YouTube. `ingest_text` takes arbitrary text. `search_documents` runs hybrid search and `query_documents` answers a question from the documents it finds. `check_processing_status` shows the progress of a background load, and `get_system_status` and `health_check` verify the link to the backend. Not every user may see every document. The `sign_in`, `start_sign_in` and `finish_sign_in` tools pass a source's credentials, and then search and answers are limited to what that user is allowed to read.
Who it is for. Developers of enterprise search and RAG over documents in Alfresco, SharePoint or S3 who want to query the knowledge base from Claude Desktop.
Good fit when
- A folder of contracts and reports goes into a knowledge graph, and you ask Claude which documents mention a specific contractor.
- Staff of a corporate repository get answers only from the files they are allowed to read.
- Wiki pages and YouTube videos on a topic are gathered into one base and searched without hopping between sites.
- For debugging extraction the server runs in HTTP mode and MCP Inspector connects to it.
- A large batch of files is loaded faster with the `skip_graph` flag, keeping only vector search.
Not a fit when
- You do not want to run the Flexible GraphRAG backend: the MCP server does nothing without it.
- You need search over a few notes in one folder: a heavy stack with a graph database is overkill.
- You need a source outside the list of 14: ingest works only with the listed sources.
Example request
Ingest the documents from the `./sample-docs/` folder and tell me the main conclusions in the research papers.MCP
- Transport
- stdio, http
- Authentication
- not required
Environment variables
FLEXIBLE_GRAPHRAG_TRUSTED_KEYsecret- A shared random key that lets the backend trust the server and answer without a user sign-in.
MCP_BROWSER_SIGN_IN- Enables browser sign-in tools for Microsoft 365, Box, Google and Azure Storage.
Limitations
The server talks to a FastAPI backend that is launched separately (`localhost:8000` by default) and takes its settings from `.env`: an LLM provider key, a graph database such as Neo4j and more. Without sign-in, search and answers cover all documents in the store. For the backend to trust the server without sign-in, a shared random key is set on both sides. HTTP mode and browser sign-in are switched on by separate settings. Source credentials travel as tool parameters, so do not expose them in shared chats.
How to disable. Remove the `flexible-graphrag` entry from `claude_desktop_config.json` and restart Claude Desktop. A package installed through pipx is removed with `pipx uninstall flexible-graphrag-mcp`. Stop the backend and the databases separately.
FAQ
How do I connect Flexible GraphRAG MCP to Claude Desktop?
Add a `flexible-graphrag` server to `claude_desktop_config.json` with the `uvx` command and the `flexible-graphrag-mcp` argument, then restart Claude Desktop. Ready configs for Windows and macOS are in the `claude-desktop-configs` folder.
What else has to run besides the MCP server?
The FastAPI backend: `uv run uvicorn main:app --host 0.0.0.0 --port 8000` in the `flexible-graphrag` folder. It needs a filled-in `.env` with an LLM provider key and a graph database connection.
How do I debug the server in MCP Inspector?
Start `flexible-graphrag-mcp --http --port 3001` and connect to `http://localhost:3001/mcp`. The repository has ready configs for stdio and HTTP modes.
How do I limit answers to a user's rights?
Pass credentials through `sign_in` or set `MCP_BROWSER_SIGN_IN=true` for browser sign-in. Sources with a sign-in passed are filtered by rights, the others stay unfiltered.
Can the HTTP endpoint be locked down?
Yes. Set `MCP_TRANSPORT_AUTH=true` and `MCP_AUTH_JWKS_URI`, and the server will require an OAuth2 token signed by your identity provider. The setting does not apply to stdio.
Discussion
No questions or reviews yet.
Related
MCP serversThe official Graphlit MCP server: an agent collects material from Slack, GitHub, mail and websites into a knowledge base and searches it
A local MCP server that turns a folder of documents into a knowledge base with hybrid search for Claude Code and Cursor
A knowledge graph with an MCP server: the agent searches concepts by meaning and sees how well each is supported.
A CLI and MCP server that extracts structured knowledge from documents: graphs, hypergraphs, and fact search with sources
InfraNodus MCP server turns any text into a knowledge graph and shows topic clusters, the gaps between them and research questions
A self-hosted RAG app with a web interface and a built-in MCP server: documents, Qdrant and answers with sources.
README in short
The package README opens with a platform overview and version 0.9.0, then gives a quick start for Windows and macOS with Claude Desktop configs. Separate sections explain three levels of permission filtering, token protection for HTTP mode and fixes for common failures.