m_flow
A bio-inspired cognitive memory engine — a new paradigm for Graph RAG.
M-flow is an MCP server that provides a bio-inspired cognitive memory engine using graph-based retrieval-augmented generation (RAG). Unlike traditional vector-similarity RAG systems, it organizes knowledge in a four-layer cone graph (Episode → Facet → FacetPoint → Entity) and retrieves information by scoring evidence paths through the graph structure rather than simple embedding proximity. The system supports multi-granularity queries, resolves coreferences at ingestion, and can optionally partition memories by recognized faces for multi-user scenarios. It exposes these capabilities via the Model Context Protocol, enabling AI assistants to build and query persistent, structured memory across conversations.
Key Features
Use Cases
- 01Building AI assistants with long-term episodic memory that recalls context from previous conversations
- 02Knowledge management systems that retrieve information through reasoning chains rather than keyword matching
- 03Multi-user AI agents that automatically partition and route memories based on facial recognition
- 04Personal AI companions that learn and apply procedural knowledge like workflows, habits, and decision rules
- 05Enterprise knowledge bases requiring precise factual recall with preserved dates, numbers, and entity relationships
- 06Research and analysis tools that need to traverse semantic relationships across documents and conversations
Related MCP Servers
View moremarkitdown
Python tool for converting files and office documents to Markdown.
firecrawl
The context API to search, scrape, and interact with the web at scale. 🔥
prompts.chat
f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy.
langflow
Langflow is a powerful tool for building and deploying AI-powered agents and workflows.
m_flow — FAQ
What is M-flow and how does it differ from regular RAG?+
M-flow is a graph-based memory engine for AI agents that retrieves information by scoring evidence paths through a structured knowledge graph rather than relying solely on vector similarity. While traditional RAG ranks chunks by embedding distance, M-flow organizes knowledge in a four-layer hierarchy and finds answers by following chains of reasoning through typed semantic edges.
How do I install the M-flow MCP server?+
You can install M-flow via pip with 'pip install mflow-ai' or clone the repository and install from source. For the MCP server specifically, navigate to the m_flow-mcp directory, run 'uv sync --dev --all-extras', then start it with 'uv run python src/server.py --transport sse'. You'll need to configure your API keys in a .env file.
Which AI clients work with M-flow's MCP server?+
M-flow provides a Model Context Protocol server that works with any MCP-compatible client, including Claude Desktop, Cursor, and other IDEs that support MCP. The server exposes memory operations (add, memorize, query) as tools that agents can invoke.
What are the prerequisites and do I need API keys?+
M-flow requires Python 3.10–3.13 and an LLM API key (supports OpenAI, Anthropic, Mistral, Groq, Ollama). You'll need to set LLM_API_KEY in your environment or .env file. For face recognition features, you also need the separate fanjing-face-recognition service and a FACE_API_KEY.
Is M-flow free to use?+
Yes, M-flow is open source under the Apache 2.0 license and free to use. However, you will incur costs from the LLM API provider you choose (OpenAI, Anthropic, etc.) based on your usage.
What databases does M-flow support for storage?+
M-flow supports multiple vector and graph databases including LanceDB, Neo4j, PostgreSQL with pgvector, ChromaDB, KùzuDB, and Pinecone. You can choose the database backend that fits your deployment needs.
How do I install m_flow?+
Open the source repository on GitHub and follow its README. m_flow is a mcp server — MCP Agents Market links you directly to the official repo.
Is m_flow free?+
m_flow is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.