mcp-memory-service

概述

mcp-memory-service是一个开源的内存后端，为LangGraph、CrewAI和AutoGen等AI代理管道提供持久化存储。它提供REST API和MCP协议访问，支持与claude.ai浏览器的Remote MCP集成。该服务具有带类型化边的知识图谱、自主内存合并和通过ONNX的本地嵌入功能。它可以自托管并支持OAuth 2.0身份验证，支持多代理场景下的内存共享和通过标签实现的代理间通信。

试试问 AI

装完之后，这里有 4 个你可以让 AI 做的事：

• store_memorymemory_store(content: string, tags: string[], conversation_id: string = null, agent_id: string = null)

  Store a memory entry with content and tags for later retrieval

  试试:Store that the API limit is 100 requests per minute试试:Remember that our project uses Python 3.11
• search_memoriesmemory_search(query: string, tags: string[], agent_id: string = null, limit: int = 10)

  Search stored memories using semantic search and filtering

  试试:Find memories about API rate limits试试:Search for project architecture decisions
• delete_memorymemory_delete(memory_id: string)

  Delete a specific memory entry by ID

  试试:Remove the outdated API limit information试试:Delete the temporary note about the server maintenance
• get_memory_graphmemory_graph_get(agent_id: string = null)

  Retrieve the knowledge graph of memories with relationships

  试试:Show me the knowledge graph of our system试试:Get all memories related to authentication
• add_memory_edgememory_graph_add_edge(source_id: string, target_id: string, relation: string)

  Add a relationship between two memory entries

  试试:Connect the API limit memory to the provider switching decision试试:Relate the bug fix to the original issue
• consolidate_memoriesmemory_consolidate(days_old: int = 30, compression: float = 0.5)

  Automatically consolidate old memories into compressed summaries

  试试:Consolidate memories older than 60 days试试:Compress memories with minimal detail loss

可对比工具

安装

# 基本安装
pip install mcp-memory-service

# 启动服务器
MCP_ALLOW_ANONYMOUS_ACCESS=true memory server --http

{
  "mcpServers": {
    "mcp-memory": {
      "command": "python",
      "args": ["-m", "mcp_memory_service.server"],
      "env": {
        "MCP_ALLOW_ANONYMOUS_ACCESS": "true",
        "MCP_STREAMABLE_HTTP_MODE": "1"
      }
    }
  }
}

MCP_STREAMABLE_HTTP_MODE=1 MCP_SSE_HOST=0.0.0.0 MCP_SSE_PORT=8765 python -m mcp_memory_service.server

mcp-memory-service 对比