Neura Market
topoteretes/cognee logo

topoteretes/cognee

Free

使用各种图和向量存储的 AI 应用和 Agents 记忆管理器，允许从 30+ 数据源摄取。 (cognee-mcp 的开发分支)

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
Visit Website GitHub (40 stars)

About topoteretes/cognee

cognee-mcp is a Model Context Protocol (MCP) server that wraps cognee's memory engine, allowing AI agents and applications to store, recall, and manage memory through a minimal API. It supports multiple transport modes (Streamable HTTP, SSE, stdio), cloud mode for connecting to Cognee Cloud, and API mode for linking to an existing Cognee FastAPI server. The server exposes three core tools: remember (store memory with optional session and permanent graph memory), recall (query memory with filters), and forget (delete datasets or owned memory). It includes integrated logging and session-aware memory, making it suitable for building persistent memory in agent workflows. The project is open source and designed to be run locally or via Docker.

Key Features

Multiple transports: Streamable HTTP (recommended for web), SSE (real-time streaming), and stdio (classic pipe)
Cloud Mode: connect to Cognee Cloud via --serve-url or COGNEE_SERVICE_URL
API Mode: connect to an already running Cognee FastAPI server
Minimal Memory API: remember, recall, and forget tools for agent memory workflows
Integrated logging: rotating file logs with console mirroring in dev mode
Session-aware memory: store fast session cache entries or permanent graph memory
Focused recall: query memory with optional session and search controls
Simple deletion: remove a dataset or all owned memory via forget tool

Pros & Cons

Pros
  • Open source with free usage under MIT license
  • Multiple transport options for flexible deployment
  • Works with cloud and self-hosted Cognee backends
  • Minimal API (remember/recall/forget) simplifies integration
  • Session-aware memory enables both temporary and permanent storage
  • Docker support for easy containerized deployment
Cons
  • Requires an OpenAI API key or alternative LLM provider for default operation
  • MCP sampling (to avoid API key) depends on host support, which is limited as of early 2026
  • No built-in authentication – security depends on the transport layer
  • Setup may need Node.js for optional UI bundle (visualize-graph.html)

Best For

Building persistent memory for AI agents that need to remember context across interactionsQuerying and managing memory from any MCP-compatible client (IDE, terminal, web)Integrating with MCP hosts that support sampling for LLM completions without an API keyDeploying memory services in local development or cloud environments with Docker

FAQ

What is cognee-mcp?
cognee-mcp is a Model Context Protocol server that exposes cognee's memory engine as tools (remember, recall, forget) for AI agents and applications.
How do I run cognee-mcp?
Clone the repo, navigate to cognee-mcp, install dependencies with uv, set your OpenAI API key in .env, and run 'python src/server.py' with optional transport flags (e.g., --transport http).
Do I need an API key to use cognee-mcp?
By default, you need an LLM_API_KEY (e.g., OpenAI). If your MCP host supports the sampling capability, you can set LLM_PROVIDER=mcp-sampling to delegate completions to the host, though host support varies.
What transport modes are available?
cognee-mcp supports Streamable HTTP (recommended for web), SSE (real-time streaming), and stdio (classic pipe, default).
Can I run cognee-mcp in Docker?
Yes, you can either build the Docker image locally or use a pre-built image. Instructions are provided in the repository.