minirag-mcp
minirag-mcp searches your documents with hybrid vector and BM25 search and runs on your computer without sending files out
- 60 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server reads every document in the given folder and their fragments enter the conversation with the model
- The `ingest_url` call reaches the network at addresses you ask it to fetch
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/minirag-mcpDetects 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 reads every document in the given folder and their fragments enter the conversation with the model
- The `ingest_url` call reaches the network at addresses you ask it to fetch
This is third-party code. Review the repository files before installing.
What it does
minirag-mcp is a local RAG server in Python for Claude Code, Cursor and Codex. You point it at a folder of documents and the server indexes them and serves hybrid search: vector similarity through fastembed with BM25 keyword ranking from LanceDB, fused with Reciprocal Rank Fusion. Exact identifiers and error codes surface next to semantic matches. Documents in 12 formats, from PDF, DOCX and XLSX to EPUB and Jupyter notebooks, are converted through markitdown once at ingest. The `find_mentions` tool lists every document with a word and reports how many documents were searched. The default model covers more than 50 languages, Russian included. The network is used only for the model download and the `ingest_url` call.
Who it is for. Developers and analysts who want to search their own documents from an agent locally.
Good fit when
- You want to answer from dozens of PDFs and spreadsheets without sending them to the cloud
- Every document with an error code or contract number has to be found
- Scanned pages are recognized locally with the `[ocr]` extra and join the search
- The agent should know which documents almost repeat each other
Not a fit when
- You need cloud search with shared team access
- There are only a few documents and plain file reading is enough
- You want model answers with no retrieval from your materials
Example request
Find every contract with the word "penalty" across my documents and tell me how many documents you searchedMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
BASE_DIRrequired- Folder of documents to index
Limitations
`uv` is required, the server starts with `uvx minirag-mcp` and takes its folder from the `BASE_DIR` variable. On first run an embedding model is downloaded and after that work is local. Indexing, search and reading local files do not use the network. The `[ocr]` and `[classify]` extras are installed separately. If `uvx` is not found on the client's `PATH`, give its full path in the configuration. Downloaded embedding and OCR models take disk space.
How to disable. Remove the entry with `claude mcp remove minirag`. The index lives in the folder set by `DB_PATH` and can be deleted by hand.
FAQ
How do I connect minirag-mcp to Claude Code?
Run `claude mcp add minirag --scope user --env BASE_DIR=/absolute/path/to/docs -- uvx minirag-mcp`. The folder path must be absolute.
Which formats are supported?
Twelve formats through markitdown: PDF, DOCX, PPTX, XLSX, HTML, CSV, EPUB, Jupyter notebooks, Markdown and plain text. Scans are recognized with the `[ocr]` extra.
What does find_mentions do?
It lists every document that contains a word or phrase and reports how many documents were searched. That makes answers to "list all", "how many" and "it is not there" complete.
What goes over the network?
Only the model download on first run and the `ingest_url` call when you ask to fetch a web page. Everything else runs on the computer.
Does it work with Russian?
Yes, the default model is multilingual and Russian and English corpora work out of the box.
Discussion
No questions or reviews yet.
Related
MCP serversA local MCP server that turns a folder of documents into a knowledge base with hybrid search for Claude Code and Cursor
MCP Local RAG
mcp-local-rag
Local search over your documents, semantic and keyword, with no keys, Docker or external database. Works offline after the first model download
On-premises RAG in containers with an MCP server: the agent searches your documents while data stays local
An MCP server gives agents search and answers over local documents using hybrid retrieval, reranking and section links.
MCP server for the Flexible GraphRAG platform: ingest from 14 data sources, hybrid search and AI answers over a knowledge graph
haiku.rag
haiku-rag
Agentic RAG over your documents with citations (page, heading) and a built-in LanceDB. The MCP server starts with haiku-rag mcp --stdio
README in short
The README presents a local RAG on fastmcp, fastembed and LanceDB as a Python analog of mcp-local-rag. Features, installation, client setup and sections on chunking, OCR and document labels follow.