Minima
On-premises RAG over your local files (PDF, DOCX, CSV, MD, TXT), queryable from Claude via MCP.
Add to your client
Copy the config for your MCP client and paste it into its config file.
git clone https://github.com/dmayboroda/minima && cd minima && docker compose -f docker-compose-mcp.yml --env-file .env up --buildPaste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"minima": {
"command": "uv",
"args": [
"--directory",
"/path_to_cloned_minima_project/mcp-server",
"run",
"minima"
]
}
}
}Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf
Before you start
- Docker and Docker Compose
- Local Python >= 3.10
- 'uv' installed (for running the MCP server)
- A .env file based on .env.sample with LOCAL_FILES_PATH, EMBEDDING_MODEL_ID, and EMBEDDING_SIZE set
- The Minima MCP containers running (docker compose -f docker-compose-mcp.yml)
About Minima
Minima turns a folder of local documents into a private, self-hosted RAG knowledge base that Claude can query through MCP. After indexing files into a local Qdrant vector store via Docker containers, you point Claude Desktop at the bundled mcp-server (run with uv), and Claude can call the minima-query tool to pull relevant context from your PDFs, spreadsheets, Word docs, markdown, CSVs, and text files. Everything stays on-premises except the LLM calls to Claude, making it suited for teams that want retrieval over sensitive local data without shipping the documents to a cloud RAG service.
Tools & capabilities (1)
minima-queryFind a context in local files (PDF, CSV, DOCX, MD, TXT). Takes a 'text' argument describing the context to search for, runs a retrieval over the indexed Qdrant vector store, and returns the relevant document context.
When to use it
- Let Claude answer questions grounded in a private folder of PDFs, contracts, or reports without uploading them to a cloud service
- Build an on-premises 'chat with your documents' workflow for sensitive or regulated data
- Search across mixed local file types (.pdf, .xls, .docx, .txt, .md, .csv) from inside Claude Desktop
- Run a fully local RAG pipeline (with Ollama or a custom OpenAI-compatible LLM) while still using Claude as the front-end via MCP
Security notes
Runs fully on-premises: the indexer and Qdrant vector store stay on your machine, so document contents are not sent to third parties beyond the LLM you connect (Claude, in MCP mode). Requires local Python >=3.10 and 'uv' installed. The MCP server reads files under LOCAL_FILES_PATH recursively, so scope that folder carefully. USER_ID/PASSWORD env vars are only for ChatGPT custom-GPT mode and are not needed for the MCP/Claude integration.
Minima FAQ
Does Minima send my documents to the cloud?
No. The indexer and the Qdrant vector store run on your own machine or cloud. In MCP mode only the retrieved context and your queries are passed to Claude as the LLM; in the fully local Ollama mode no external LLM is involved at all.
How do I connect Minima to Claude Desktop?
Run the MCP containers (docker compose -f docker-compose-mcp.yml --env-file .env up --build), then add the `minima` entry shown in the README to claude_desktop_config.json, pointing the `uv --directory` arg at the cloned project's mcp-server folder. Restart Claude and ask it to find context in your local files.
What file types can it index?
Indexing is recursive over LOCAL_FILES_PATH and supports .pdf, .xls, .docx, .txt, .md, and .csv files.
Can I use it with GitHub Copilot instead of Claude?
Yes. The README documents a `.vscode/mcp.json` configuration that runs the bundled `run_in_copilot.sh` script as a stdio MCP server for Copilot.
Which tools does the MCP server expose?
A single tool, `minima-query`, which retrieves relevant context from your indexed local files based on a search string.
Alternatives to Minima
Compare all alternatives →
Official MCP server providing persistent, file-backed knowledge-graph memory across sessions.
Structured step-by-step reasoning tool for breaking problems into revisable thought sequences.
Fully managed remote server for AWS docs, blogs, What's-New and Well-Architected guidance — no key.