@memly/mcp-server

Persistent memory for IDEs that don't support custom LLM providers.

Proxy vs MCP — which one should I use?

Memly has two integration modes:

| Mode | How it works | Best for | |------|-------------|----------| | Proxy (transparent) | Point your IDE's base URL to api.memly.site — memory injected in 100% of requests, invisible | IDEs that support a custom OpenAI-compatible endpoint | | MCP server (this package) | Memly runs as an MCP tool — load_context is called automatically at session start | IDEs locked to their own AI service |

Which mode for my IDE?

✅ Use the Proxy — supports custom endpoint

| IDE / Tool | Where to configure | |-----------|-------------------| | Cursor | Settings → Models → OpenAI Base URL | | Continue.dev (VS Code / JetBrains extension) | config.json → apiBase | | Zed | settings.json → api_url | | Aider | --openai-api-base CLI flag | | Jan.ai | Settings → Model → Engine URL | | LM Studio | Server settings | | Open WebUI | Admin → Connections | | Msty | Settings → Custom provider | | Void | Settings → Custom provider |

❌ Use the MCP server — locked provider, no custom endpoint

| IDE / Tool | Reason | |-----------|--------| | VS Code + GitHub Copilot | Microsoft/GitHub auth — base URL not configurable | | JetBrains AI Assistant | Locked to JetBrains AI Service subscription | | Windsurf (Codeium) | Cascade uses Codeium's own models — no external endpoint | | Amazon Q / CodeWhisperer | AWS-only pipeline | | Tabnine | Proprietary closed service | | Replit AI | Built into Replit, no external provider | | Gitpod AI | Locked to their own service | | Sourcegraph Cody | Locked on free tier | | Claude Desktop | Chat app, not an IDE — MCP native |

Quick Start

1. Get your API key

Go to memly.site/dashboard/api-keys

2. Configure your IDE's MCP server

VS Code / GitHub Copilot — create .vscode/mcp.json in your project:

{
  "servers": {
    "memly": {
      "command": "npx",
      "args": ["-y", "@memly/mcp-server"],
      "env": {
        "MEMLY_API_KEY": "memly_your_key_here"
      }
    }
  }
}

JetBrains (IntelliJ, WebStorm, PyCharm…) — create .idea/mcp.json:

{
  "servers": {
    "memly": {
      "command": "npx",
      "args": ["-y", "@memly/mcp-server"],
      "env": {
        "MEMLY_API_KEY": "memly_your_key_here"
      }
    }
  }
}

Windsurf — Settings → Cascade → MCP Servers → Add:

• Command: npx
• Args: -y @memly/mcp-server
• Env: MEMLY_API_KEY=memly_your_key_here

Claude Desktop — edit claude_desktop_config.json:

{
  "mcpServers": {
    "memly": {
      "command": "npx",
      "args": ["-y", "@memly/mcp-server"],
      "env": {
        "MEMLY_API_KEY": "memly_your_key_here"
      }
    }
  }
}

3. Run auto-setup once per project

Writes the instruction file for your IDE so load_context runs automatically at every session start:

npx @memly/mcp-server init

Auto-detects VS Code, Cursor, Windsurf, Claude Desktop. Run once, never think about it again.

Tools

| Tool | When the AI calls it | |------|---------------------| | load_context | Automatically at session start — loads memories from previous sessions | | search_memories | When you ask about something specific not loaded by load_context | | remember | When you make a decision, solve a problem, or say "remember this" | | list_projects | When you ask to list or switch projects |

Environment Variables

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | MEMLY_API_KEY | ✅ | — | Your Memly API key (memly_...) | | MEMLY_API_URL | — | https://api.memly.site | Override for self-hosted deployments | | MEMLY_PORT | — | 3800 | HTTP transport port (--http mode only) |

Self-Hosted

Run alongside your Memly proxy on your VPS for zero-latency:

MEMLY_API_KEY=memly_... bun run packages/mcp-server/src/index.ts --http

License

BSL-1.1 — Memly Community Edition