habage-mcp

habage-mcp is a stdio MCP bridge for Habage. It lets local AI clients such as Claude Desktop, Claude Code-style tools, Codex, and Gemini-compatible MCP clients talk to your signed-in Habage workspace through the app's authenticated /mcp endpoint.

The package does not connect to MongoDB and does not create board tasks directly. Habage keeps auth, proposal review, task creation, and user ownership inside the web app.

This package is for local stdio MCP clients that can run a command such as npx habage-mcp. For ChatGPT custom connectors, use your deployed Habage remote MCP URL: https://habage.vercel.app/mcp.

Requirements

• Node.js 20.10 or newer
• A reachable Habage app, usually https://habage.vercel.app/
• A user-owned MCP token from Habage at /dashboard/mcp-tools

Environment

HABAGE_MCP_URL=https://habage.vercel.app/mcp
HABAGE_MCP_TOKEN=<token-from-dashboard-mcp-tools>

HABAGE_MCP_URL is optional and defaults to https://habage.vercel.app/mcp.

HABAGE_MCP_TOKEN is required. It scopes all MCP writes to the user who created the token.

MCP Client Config

For local development before publishing the package, build it and point your MCP client at the compiled bridge:

bun run --cwd packages/habage-mcp build

{
  "mcpServers": {
    "habage": {
      "command": "node",
      "args": [
        "C:\\Users\\prath\\OneDrive\\Desktop\\study\\temp\\open-source\\habage\\packages\\habage-mcp\\dist\\stdio-proxy.js"
      ],
      "env": {
        "HABAGE_MCP_URL": "https://habage.vercel.app/mcp",
        "HABAGE_MCP_TOKEN": "<token-from-dashboard-mcp-tools>"
      }
    }
  }
}

After publishing, the same server can be configured through npx:

{
  "mcpServers": {
    "habage": {
      "command": "npx",
      "args": ["habage-mcp"],
      "env": {
        "HABAGE_MCP_URL": "https://habage.vercel.app/mcp",
        "HABAGE_MCP_TOKEN": "<token-from-dashboard-mcp-tools>"
      }
    }
  }
}

On native Windows, some clients need the command wrapped as cmd /c npx -y habage-mcp instead of calling npx directly.

Publishing

npm login
bun run --cwd packages/habage-mcp build
cd packages/habage-mcp
npm pack --dry-run
npm publish

Recommended Proposal Flow

1. Call prepare_task_breakdown_prompt for weak or unclear prompts.
2. Ask the returned clarification questions when needed.
3. Generate the proposal tasks in the AI client.
4. Submit them with propose_task_breakdown_from_tasks.
5. Review the proposal in Habage and add selected tasks to Todo on demand.