MemTrace MCP Server

MCP (Model Context Protocol) server for MemTrace - an AI context memory service that helps AI assistants remember and retrieve information across conversations.

Features

• 🔍 Semantic Search: Find relevant knowledge using natural language queries
• 💾 Knowledge Storage: Store important information, decisions, and insights
• 🧠 LLM-Powered Extraction: Automatically extract structured knowledge from text
• 🔄 Context Retrieval: Get startup context and pull relevant knowledge
• 🏷️ Categorization: Organize knowledge by categories (facts, preferences, decisions, etc.)

Installation

Option 1: npx (Recommended - Always Latest)

No installation needed! Just configure your MCP client:

{
  "mcpServers": {
    "memtrace": {
      "command": "npx",
      "args": ["-y", "memtrace-mcp-server"],
      "env": {
        "MEMTRACE_API_URL": "https://your-memtrace-server.com/api/v1",
        "MEMTRACE_TOKEN": "your-token-here"
      }
    }
  }
}

Option 2: Global Installation

npm install -g memtrace-mcp-server

Then configure:

{
  "mcpServers": {
    "memtrace": {
      "command": "memtrace-mcp-server",
      "env": {
        "MEMTRACE_API_URL": "https://your-memtrace-server.com/api/v1",
        "MEMTRACE_TOKEN": "your-token-here"
      }
    }
  }
}

Configuration

Required Environment Variables

• MEMTRACE_TOKEN: Your MemTrace API token (get it from your MemTrace dashboard)
• MEMTRACE_API_URL: MemTrace API base URL (default: http://localhost:3000/api/v1)

Getting Your API Token

1. Log in to your MemTrace instance
2. Go to Dashboard → Tokens
3. Create a new MCP token with required permissions
4. Copy the token and add it to your configuration

Available Tools

search_knowledge

Search for relevant knowledge using semantic search.

search_knowledge({
  query: "search query here",
  limit: 10  // optional, default: 10
})

push_knowledge

Store knowledge by submitting raw content. The server will automatically extract structured knowledge using LLM.

Important: It's recommended to summarize/extract key points on the client side before pushing for better results.

push_knowledge({
  content: "Raw text content containing important information...",
  project_id: "optional-project-id"  // optional
})

Content Size Limits (per tier):

• Free: 10,000 characters
• Pro: 50,000 characters
• Business: 100,000 characters

pull_knowledge

Retrieve recent knowledge items.

pull_knowledge({
  limit: 10,  // optional, default: 10
  category: "fact"  // optional: fact, preference, procedure, decision, insight, other
})

extract_knowledge

Extract knowledge from text/conversation using LLM (preview mode, doesn't save unless save: true).

extract_knowledge({
  content: "Text to extract knowledge from...",
  save: false,  // optional, default: false
  project_id: "optional-project-id"  // optional
})

get_startup_context

Get startup context with pinned knowledge and recent items for project initialization.

get_startup_context({
  project_id: "optional-project-id",  // optional
  max_tokens: 4000  // optional, default: 4000
})

Usage Examples

With Claude Code

1. Install Claude Code CLI
2. Add MemTrace MCP server to ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "memtrace": {
      "command": "npx",
      "args": ["-y", "memtrace-mcp-server"],
      "env": {
        "MEMTRACE_API_URL": "https://your-memtrace-server.com/api/v1",
        "MEMTRACE_TOKEN": "your-mcp-token-here"
      }
    }
  }
}

3. Restart Claude Code

With Cursor

1. Add to Cursor MCP configuration
2. Restart Cursor

Rate Limits

API rate limits vary by subscription tier:

| Tier | Daily Limit | Burst (per minute) | |------|-------------|-------------------| | Free | 50 requests/day | 10 requests/min | | Pro | 200 requests/day | 30 requests/min | | Business | 1000 requests/day | 100 requests/min |

Troubleshooting

"MEMTRACE_TOKEN environment variable is required"

Make sure you've set the MEMTRACE_TOKEN in your MCP server configuration.

"Failed to push knowledge"

Check:

1. Your API token is valid
2. Content doesn't exceed tier limits
3. Server logs for detailed error messages

Updating to Latest Version

If using npx -y:

• Just restart your MCP client (Claude Code / Cursor)
• It will automatically fetch the latest version

If using global installation:

npm update -g memtrace-mcp-server

License

MIT