cognee-mcp-server

by MCP-Mirror

304 downloads
Not rated
GitHub

About

cognee-mcp-server is an MCP server for cognee, an AI memory engine. It provides a tool that builds a knowledge graph from input text and performs search within that graph, intended for developers integrating memory capabilities into AI assistants.

Details

Author
MCP-Mirror
Downloads
304
Categories
Other

- Builds knowledge graphs from input text
- Performs search within constructed knowledge graphs
- Supports custom Pydantic graph models
- Configurable with Claude Desktop via uvx
- Uses multiple database providers (NetworkX, LanceDB, SQLite)

Install with uv (Python package manager) and configure in Claude Desktop by adding an entry to claude_desktop_config.json using uvx. Invoke the Cognify_and_search tool with required inputs text and search_query, and optional graph_model_file and graph_model_name for a custom Pydantic graph model.

cognee-mcp-server

An MCP server for cognee, an AI memory engine.

Tools

- Cognify_and_search : Builds knowledge graph from the input text and performs search in it.
- Inputs:
- text (String): Context for knowledge graph contstruction
- search_query (String): Query for retrieval
- graph_model_file (String, optional): Filename of a custom pydantic graph model implementation
- graph_model_name (String, optional): Class name of a custom pydantic graph model implementation
- Output:
- Retrieved edges of the knowledge graph

Configuration

Usage with Claude Desktop

Install uv with homebrew.

Add this to your claude_desktop_config.json:
<details>
<summary>Using uvx</summary>

"mcpcognee": {
  "command": "uv",
  "args": [
    "--directory",
    "/path/to/your/cognee-mcp-server",
    "run",
    "mcpcognee"
  ],
  "env": {
    "ENV": "local",
    "TOKENIZERS_PARALLELISM": "false",
    "LLM_API_KEY": “your llm api key”,
    "GRAPH_DATABASE_PROVIDER": “networkx”,
    "VECTOR_DB_PROVIDER": "lancedb",
    "DB_PROVIDER": "sqlite",
    "DB_NAME": “cognee_db”
  }
}
</details>
No reviews yet — be the first

Sign in to leave a review

Use Google, GitHub, or an email account so ratings stay tied to real people.

Email sign in

No reviews posted yet.