Minima

Minima is an MCP server that enables AI clients to search and retrieve context from local documents using an on-premises retrieval-augmented generation engine. The platform connects host systems and AI applications—including Claude Desktop and GitHub Copilot—to files stored on a local computer or private server. It targets developers, data privacy teams, and researchers who require retrieval workflows without exposing sensitive internal documents to third-party services. Minima indexes directories containing various formats like PDF, DOCX, TXT, CSV, and Markdown into a vector database, using Sentence Transformer embedding models to capture semantic relationships. When integrated as an MCP server or paired with local or custom inference engines, Minima provides document-based question answering directly from workspace directories. Users can operate it in an isolated mode with Ollama, connect to custom OpenAI-compatible inference backends such as vLLM or LiteLLM, or bind it directly to desktop AI applications. By executing indexing and semantic retrieval within private containers, Minima provides local knowledge grounding while giving users complete control over their model infrastructure and file privacy.

Category: AI Memory & Context

Tags: embeddings, local-llm, rag, vector-search

Visit Minima

How to install and configure Minima

1. Confirm your environment has Python 3.10 or higher and uv installed, then clone the repository at https://github.com/dmayboroda/minima. 2. Create a .env file in the project root by copying .env.sample, and configure the LOCAL_FILES_PATH, EMBEDDING_MODEL_ID, and EMBEDDING_SIZE variables. 3. Start the MCP server container by running: docker compose -f docker-compose-mcp.yml --env-file .env up --build. 4. Register Minima in your Claude Desktop configuration file (claude_desktop_config.json) under mcpServers: {"mcpServers": {"minima": {"command": "uv", "args": ["--directory", "/path_to_cloned_minima_project/mcp-server", "run", "minima"]}}}. 5. Restart Claude Desktop to enable local file retrieval.

What you can do with Minima

• Searching local documentation collections, such as PDF user manuals and Markdown guides, to answer engineering queries directly inside Claude Desktop. * Performing semantic searches across proprietary CSV and spreadsheet datasets without transmitting sensitive corporate financial data to third-party cloud APIs. * Grounding developer assistants in local code repositories and technical writeups via GitHub Copilot using the stdio MCP configuration. * Querying private enterprise folders containing DOCX and text documents using a custom self-hosted LLM backend like vLLM or Ollama.

Key facts

• Open Source
• https://github.com/dmayboroda/minima
• AI Memory & Context, Notes & Knowledge Management
• embeddings, local-llm, rag, vector-search

Part of MCP Servers

How do I install Minima?

To install Minima, clone the project repository and create a .env file from the provided template. Configure the path to your local files and your chosen Sentence Transformer embedding model. Make sure your system has Python 3.10 or higher and uv installed, then build and launch the container with Docker Compose using the MCP profile. You can then reference the server in your MCP client.

What can Minima do?

Minima provides retrieval-augmented generation across locally hosted directories. It recursively indexes documents such as PDF, Word, plain text, Markdown, and spreadsheet files using vector embeddings stored in Qdrant. AI clients can perform semantic searches and query local files directly without uploading documents to external cloud infrastructure, supporting fully isolated on-premises setups or custom model endpoints.

Which MCP clients work with Minima?

Minima works with Anthropic Claude Desktop and GitHub Copilot in Visual Studio Code through Model Context Protocol configurations. Users can register Minima in Claude Desktop using standard uv commands or configure it as a stdio server in VS Code settings. Minima also supports custom OpenAI-compatible endpoints, an Electron interface, and custom ChatGPT integrations.

Is Minima open source?

Yes, Minima is an open source project distributed under the Mozilla Public License version 2.0. You can inspect the source code, review Docker deployment configurations, and customize the embedding or indexing logic directly within the GitHub repository without licensing fees.