Task Manager MCP Server

What is Task Manager MCP Server?

A Model Context Protocol (MCP) server that integrates with AI code editors like Cursor to provide complex feature planning, task management, and code review. It uses Gemini 2.5's 1‑million‑token context window and a Svelte UI to overcome Cursor's limited context, enabling unlimited context and usage for the price of Cursor's subscription.

How to use Task Manager MCP Server?

After cloning and building (npm install && npm run build), configure the server in Cursor’s mcp.json with the absolute path to dist/server.js and your API keys. Create a custom Cursor mode with tools like plan_feature, mark_task_complete, and get_next_task enabled, then paste the recommended instructions (e.g., “Always use plan_feature first, complete one task at a time”). In that mode, give Cursor a feature request; it will call plan_feature, return the first task, and you repeat the workflow until done.

Key features of Task Manager MCP Server

• Plans complex features step‑by‑step using an LLM with repomix project context.
• Provides a Svelte UI to view tasks, adjust plans, and review changes.
• Uses Gemini 2.5’s 1‑million‑token context window with automatic truncation.
• Stores conversation history per feature in .mcp/features/ JSON files.
• Supports clarification workflow via WebSocket when the LLM needs more info.
• Automatically runs code review on git diff HEAD after the last task (optional).

Use cases of Task Manager MCP Server

• Implementing large features in Cursor without exhausting the context window.
• Breaking down high‑effort coding tasks into recursive, manageable steps.
• Allowing users to adjust a plan mid‑development through the UI.
• Automatically reviewing code changes and generating new tasks if needed.

FAQ from Task Manager MCP Server

What is the main purpose of this server?

It lets Cursor (or similar AI editors) plan and execute complex features one task at a time, overcoming context‑window limits by using Gemini 2.5’s 1‑million‑token context and a structured workflow.

Does it require Cursor?

The README states Cursor is the primary intended editor, but the server may work with other AI code editors; it has not been tested elsewhere.

Which LLMs are supported?

It uses OpenRouter (recommended) or Google AI API. Default models are google/gemini-2.5-flash-preview:thinking with a fallback to the same model. Users can configure different models in environment variables.

How does the server handle context and costs?

It leverages Gemini 2.5’s large context via OpenRouter’s free tiers to minimize costs. Rate‑limited requests automatically fall back to a cheaper model. For unlimited usage, the recommended setup is Cursor’s $20/month subscription plus your own Google AI API key.

What data is stored and where?

Conversation history and task data are stored in JSON files under .mcp/features/ inside the project, one folder per feature. The UI server reads this data to display progress.