Stop losing context when you switch between AI coding tools.

A Model Context Protocol (MCP) server that lets Claude Code, Cursor, Codex, and other AI coding assistants share logs and session history with each other.

You're coding with Claude Code. You make progress. Then you switch to Cursor to test something. Now you've lost all your context. You explain everything again. Then you jump to Codex. Explain it all over again.

It's exhausting.

Colab MCP is a shared MCP server that exposes your chat logs, terminal history, and IDE events as tools and resources across all your AI coding assistants.

When you switch tools, your AI already knows what you were working on. No more copy-pasting. No more re-explaining. Just continuous flow.

• 🔄 Share context across tools - Claude Code, Cursor, Codex, Gemini
• 📜 Access chat transcripts from previous sessions
• 🔍 Search across all logs - find that conversation from last week
• 🎯 Session summaries - quick overview of what you were working on
• 🖥️ Terminal & IDE event tracking - see what commands were run
• 🚀 Fast setup - one command to install across all your tools

pip install colab-mcp

Run the interactive installer:

sudo colab-mcp-install

The installer will:

• 🔍 Detect which AI coding tools you have installed
• ✅ Let you choose which ones to configure
• ⚙️ Add Colab MCP to their MCP server configs
• 📝 Give you instructions to restart each tool

Restart Claude Code, Cursor, Codex, or whichever tools you configured.

That's it! 🎉

Once installed, Colab MCP exposes several tools and resources to your AI assistants:

• list_sessions - Get a list of all coding sessions
• fetch_transcript - Retrieve the full transcript of a session
• summarize_session - Get a quick summary of what happened
• search_logs - Search across all logs (chat, MCP, IDE events)
• codex_status - Check recent Codex CLI activity

Try asking your AI assistant:

"What was I working on in my last session?"

"Search my logs for discussions about authentication"

"Summarize my session from yesterday afternoon"

"What errors did I encounter in the last hour?"

If you prefer to configure manually, add this to your MCP config:

{
  "servers": {
    "colab-mcp": {
      "command": "colab-mcp",
      "env": {
        "CLAUDE_HOME": "/home/yourusername/.claude",
        "CURSOR_LOGS": "/home/yourusername/.cursor-server/data/logs",
        "TMPDIR": "/tmp"
      }
    }
  }
}

{
  "mcpServers": {
    "colab-mcp": {
      "command": "colab-mcp",
      "env": {
        "CLAUDE_HOME": "/home/yourusername/.claude",
        "CURSOR_LOGS": "/home/yourusername/.cursor-server/data/logs",
        "TMPDIR": "/tmp"
      }
    }
  }
}

[mcp_servers.colab-mcp]
command = "colab-mcp"
args = []
env = { CLAUDE_HOME = "/home/yourusername/.claude", CURSOR_LOGS = "/home/yourusername/.cursor-server/data/logs", TMPDIR = "/tmp" }

graph TB
    subgraph AI["AI Tools"]
        Claude[Claude Code]
        Cursor[Cursor]
        Codex[Codex]
    end
    
    MCP[Colab MCP Server]
    
    subgraph Logs["Log Files"]
        Chat[Chat History]
        IDE[IDE Events]
        Term[Terminal]
    end
    
    Claude --> MCP
    Cursor --> MCP
    Codex --> MCP
    
    MCP --> Chat
    MCP --> IDE
    MCP --> Term
    
    style MCP fill:#e8f4f8,stroke:#4a90a4,stroke-width:2px
    style AI fill:#f9f9f9,stroke:#ccc
    style Logs fill:#f9f9f9,stroke:#ccc

Contributions are welcome! Check out the docs/ folder for more detailed information about how Colab MCP works.

MIT License - see LICENSE for details.

Built with FastMCP - the fastest way to build MCP servers in Python.

Made with ❤️ by developers tired of losing context