agentdb-mcp

MCP server for AgentDB. Gives any MCP-compatible AI agent persistent memory, semantic search, causal debugging, and time travel — no SDK, no code changes.

Works with Claude Desktop, Cursor, Windsurf, Zed, LangChain MCP, CrewAI, AutoGen, and any framework that supports the Model Context Protocol.

Install

pip install agentdb-mcp

Or run directly with uvx (no install needed):

uvx agentdb-mcp

Quickstart

Get an API key at agentdb.zizka.ai, then add AgentDB to your MCP config.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "agentdb": {
      "command": "uvx",
      "args": ["agentdb-mcp"],
      "env": {
        "AGENTDB_API_KEY": "agdb_live_xxxx"
      }
    }
  }
}

Restart Claude Desktop. You'll see AgentDB tools appear in the tool list.

Cursor

Edit your MCP config at ~/.cursor/mcp.json:

{
  "mcpServers": {
    "agentdb": {
      "command": "uvx",
      "args": ["agentdb-mcp"],
      "env": {
        "AGENTDB_API_KEY": "agdb_live_xxxx"
      }
    }
  }
}

Self-hosted AgentDB

{
  "mcpServers": {
    "agentdb": {
      "command": "uvx",
      "args": ["agentdb-mcp"],
      "env": {
        "AGENTDB_HOST": "http://localhost:8000",
        "AGENTDB_API_KEY": ""
      }
    }
  }
}

Available Tools

Example: Claude with persistent memory

Once configured, Claude can call AgentDB tools in any conversation:

"Remember that the user prefers short replies" → Claude calls log_event("assistant", "preference", {"note": "user prefers short replies", "user_id": "u_123"})

"What do I know about this user?" → Claude calls search_memory("user preferences and history", agent="assistant")

"What did you tell this user last week?" → Claude calls time_travel("assistant", "2026-05-03T00:00:00Z")

Environment Variables

License

MIT