Minima
On-premises RAG in containers with an MCP server: the agent searches your documents while data stays local
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- Indexes local documents in containers
- In some modes sends queries to external models
Install
Manual install
docker compose -f docker-compose-mcp.yml --env-file .env up --buildFill in .env from .env.sample before launching. Other modes have their own compose files.
This is third-party code. Review the repository files before installing.
What it does
Minima is an open-source RAG that runs in containers and indexes local documents. It works in several modes: fully local with Ollama, with your own model over an OpenAI-compatible API, integrated into ChatGPT via a custom GPT, and with the Anthropic Claude app. In MCP mode the Claude Desktop app connects to the Minima server and searches your files while the indexer stays on your machine. It launches through docker compose with a separate file per mode and environment variables in .env.
Who it is for. For teams and specialists who need search over their own documents with data kept on their own infrastructure.
Good fit when
- You need to search local documents through Claude or ChatGPT
- You need a RAG where data does not leave your side
- You want to connect a local knowledge base to an agent via MCP
Not a fit when
- You cannot run Docker containers
- You want a ready cloud search without your own infrastructure
Example request
Search my local documents for the sections on the leave policy and cite the sourcesLimitations
It needs Docker and .env setup, and the local mode needs Ollama and reranker models. The ChatGPT and Claude modes assume access to external services, some of which do not work from Russia without VPN. Deployment and maintenance are on the user.
How to disable. Stop the containers with docker compose down and remove the minima server from the MCP config.
MCP
- Transport
- stdio
- Authentication
- not required
| Environment variables | |
|---|---|
| LOCAL_FILES_PATH required | Path to the folder of documents to index |
| EMBEDDING_MODEL_ID required | Embedding model for indexing |
| LLM_API_KEY secret | API key for your custom model if it requires authorization |
Security check
- Indexes local documents in containers
- In some modes sends queries to external models
README in short
The README describes an open-source containerized RAG that indexes local documents and runs in four modes: local on Ollama, your own model over an OpenAI-compatible API, ChatGPT via a custom GPT, and the Claude app. It launches through docker compose with a file per mode and .env variables, and for MCP it shows a config snippet that runs via uv. There is a web UI and an Electron desktop app. MPL-2.0 licensed.
FAQ
Does data leave my side?
In the local Ollama mode everything runs on your machine. In the ChatGPT and Claude modes the primary model is external while the indexer stays local.
Which modes are there?
Fully local on Ollama, your own model over an OpenAI-compatible API, ChatGPT integration, and the Claude app via MCP.
Related
166 skills for scientific work: bioinformatics, cheminformatics, clinical data, geospatial analysis and 100+ databases
Google's open-source MCP server for databases: ready tools for Postgres, MySQL, BigQuery, Spanner and more, plus custom tools in tools.yaml
Official Hugging Face skills: Hub operations via the hf CLI, datasets, model training, Spaces, evals and deployment
A visualization language for agents and an MCP server: neat charts from a simple semantic spec