MCP Server for MemOS API

A Model Context Protocol (MCP) implementation for the MemOS API service. This package provides a command-line interface to interact with MemOS API using MCP standards.

MCP Configuration

To use this package In MCP Client, use the following configuration: （You may need to install Node.js first）

{
  "mcpServers": {
    "memos-api-mcp": {
      "command": "npx",
      "args": ["-y", "@memtensor/memos-api-mcp"],
      "env": {
        "MEMOS_API_KEY": "your-api-key",
        "MEMOS_USER_ID": "your-user-id",
        "MEMOS_CHANNEL": "the-site-where-you-are-seeing-this-document"
      }
    }
  }
}

Configuration Explanation

• command: Uses npx to run the package
• args: Arguments passed to npx to run the package
• env: Environment variables
  • MEMOS_API_KEY: Your Memos API key for authentication (Get your API Key in Dashboard[https://memos-dashboard-pre.openmem.net/cn/quickstart/])
  • MEMOS_USER_ID: Stable per-human identifier. MUST be deterministic and non-PII, and MUST remain the same for the same person across devices/sessions. NEVER reuse across different people. DO NOT use random values, device IDs, or model/chat session IDs. Recommended: SHA-256(lowercase(trim(email))) or your SSO subject/employee ID.
  • MEMOS_CHANNEL: The site where you are seeing this document. Candidate values: MODELSCOPE, MCPSO, MCPMARKETCN, MCPMARKETCOM, GLAMA, PULSEMCP, MCPSERVERS,LOBEHUB,MEMOS(meaning the MemOS official website), GITHUB(meaning this README file)

Available MCP Tools

This package provides the following MCP tools:

1. add_message

   • Adds a new message to a conversation
   • Parameters:
     • conversation_id: Unique identifier of the conversation associated with the feedback.
     • messages: Array of messages containing role and content information.
       • role: Role of the message sender (user or assistant).
       • content: Message content.
       • chat_time: (Optional) Message timestamp.

2. search_memory

   • Searches for memories in a conversation.
   • Parameters:
     • query: Text content to search within the memories. The token limit for a single query is 4k.
     • filter: (Optional) Filter conditions, used to precisely limit the memory scope before retrieval.
     • knowledgebase_ids: (Optional) Array specifying the knowledge bases to search.
       • DO NOT USE THIS unless the user explicitly mentions "knowledge base" or "KB".
       • 1. If the user explicitly asks to search ALL knowledge bases -> pass ["all"].
       • 2. If the user specifies particular KB IDs -> pass those IDs.
       • 3. If the user DOES NOT mention knowledge bases -> OMIT this parameter (do not send it).
     • include_preference: (Optional) Enable preference memory recall. Default: true.
     • preference_limit_number: (Optional) Max preference memories to return. Default: 9, max 25.
     • include_tool_memory: (Optional) Enable tool memory recall. Default: false.
     • tool_memory_limit_number: (Optional) Max tool memories to return. Default: 6, max 25.
     • include_skill: (Optional) Enable Skill recall. Default: false.
     • skill_limit_number: (Optional) Max Skills to return. Default: 6, max 25.
     • relativity: (Optional) Relevance threshold (0-1) for recalled memories. A value of 0 disables relevance filtering.
     • conversation_first_message: First user message in the thread (used to generate conversation_id).
     • memory_limit_number: Maximum number of memories that can be recalled. Default: 9, max 25.

3. delete_memory

   • Delete specific memories by their IDs.
   • Parameters:
     • user_ids: List of user IDs whose memories will be deleted.
     • memory_ids: List of memory IDs to delete.

4. add_feedback

   • Submit user feedback to the MemOS system.
   • Note: Feedback is applied asynchronously — add_feedback returns immediately (often with a task_id), and the effect may take a short time to appear.
   • Parameters:
     • user_id: The user identifier associated with the feedback.
     • conversation_id: Unique identifier of the conversation associated with the feedback.
     • feedback_content: The specific content of the feedback.
     • agent_id: (Optional) Agent ID associated with the feedback.
     • app_id: (Optional) App ID associated with the feedback.
     • feedback_time: (Optional) Feedback time string (default: current UTC time).
     • allow_public: (Optional) Whether to allow public access (default: false).
     • allow_knowledgebase_ids: (Optional) List of knowledge base IDs allowed to be written to.

All tools use the same configuration and require the MEMOS_API_KEY environment variable.

Features

• MCP-compliant API interface
• Command-line tool for easy interaction
• Built with TypeScript for type safety
• Express.js server implementation
• Zod schema validation

Prerequisites

• Node.js >= 18
• npm or pnpm (recommended)

Installation

You can install the package globally using npm:

npm install -g @memtensor/memos-api-mcp

Or using pnpm:

pnpm add -g @memtensor/memos-api-mcp

Usage

After installation, you can run the CLI tool using:

npx @memtensor/memos-api-mcp

Or if installed globally:

memos-api-mcp

Development

1. Clone the repository:

git clone <repository-url>
cd memos-api-mcp

2. Install dependencies:

pnpm install

3. Start development server:

pnpm dev

4. Build the project:

pnpm build

Available Scripts

• pnpm build - Build the project
• pnpm dev - Start development server using tsx
• pnpm start - Run the built version
• pnpm inspect - Inspect the MCP implementation using @modelcontextprotocol/inspector

Project Structure

memos-mcp/
├── src/           # Source code
├── build/         # Compiled JavaScript files
├── package.json   # Project configuration
└── tsconfig.json  # TypeScript configuration

Dependencies

• @modelcontextprotocol/sdk: ^1.0.0
• express: ^4.19.2
• zod: ^3.23.8
• ts-md5: ^2.0.0

Version

Current version: 1.0.0-beta.2