Skip to main content
Glama

task-agent-mcp

An MCP server that takes a coding goal for a Python project, asks Claude to split it into subtasks, has Claude rewrite the relevant files for each one, and runs pytest after every subtask. If the tests pass it commits, otherwise it puts the files back.

Install and run

git clone https://github.com/yerramsettianish0427/task-agent-mcp.git
cd task-agent-mcp
pip install -r requirements.txt
export ANTHROPIC_API_KEY=...
python task_agent_mcp.py

It talks MCP over stdio, so normally a client starts it. For Claude Desktop, add this to claude_desktop_config.json:

{
  "mcpServers": {
    "task-agent": {
      "command": "python",
      "args": ["/absolute/path/to/task_agent_mcp.py"],
      "env": { "ANTHROPIC_API_KEY": "..." }
    }
  }
}

The tools are scan_project, plan_task (subtasks only, no edits), execute_task, check_status and rollback_all.

Related MCP server: ralph-loop-mcp

Example

Listing the tools from an MCP client and calling check_status before any task has run:

tools: ['scan_project', 'plan_task', 'execute_task', 'check_status', 'rollback_all']
check_status: {"status": "idle", "message": "No active task"}

execute_task returns JSON with each subtask, the files it touched, whether the tests passed, and whether it was committed, rolled back or skipped.

Limits

  • The target project needs a git repo and a pytest suite. If pytest fails or finds no tests, every subtask gets rolled back.

  • Planning only sees the first 800 characters of the five largest files, and Claude returns whole files rather than diffs.

  • Python files only, one task at a time, and execute_task blocks until it finishes, so check_status is mostly useful afterwards.

  • rollback_all discards uncommitted changes. It doesn't undo commits.

  • The model is hard-coded (claude-sonnet-5-5).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      An autonomous multi-agent orchestration layer that generates a bespoke software engineering organization for a given objective, runs parallel OpenCode workers in isolated git worktrees, and integrates reviewed code through a replan loop.
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      A machine-enforced workflow protocol for AI coding agents that eliminates false progress. Tasks require verifiable evidence to close, while decisions, assumptions, and open questions persist in a versioned reasoning graph for seamless agent handoffs. Designed to keep deep, long-running projects coherent across sessions, agents, and months.
      39
      32
      AGPL 3.0