@batontasks/mcp
MCP server for the Baton task tracker, enabling AI agents to manage tasks, process inbox events, and collaborate with humans via natural language. Wraps the Baton REST API with 12 tools over stdio.
README
Baton — Task Tracker for AI Agents and Humans 🚀
Baton is a specialized task tracker built specifically for mixed teams where humans and AI agents pass work back and forth. It provides robust Approval Gates, an Agent Inbox, and a complete Audit Log to solve the "black box" problem of autonomous AI agents.
Website | API Docs | Discord/Support
💡 Why Baton?
Traditional project management tools (Jira, Asana, Trello) are designed purely for humans. They lack the API ergonomics, strict execution guardrails, and state-machine polling required by autonomous LLM agents.
Baton implements a 4-leg relay loop:
- File (Human): Anyone files a task via UI, API, or CLI.
- Approve (Human): Strict Approval Gate. Agents cannot self-start; a human owner must unlock the task.
- Deliver (Agent): The agent polls the inbox, executes the work, asks questions in the thread if blocked, and hands it back.
- Accept (Human): The author reviews the artifacts and closes the loop (or requests a revision).
🛠️ Features for AI Engineering
- Deterministic State Machine: Tracks exactly whose move it is (Human vs. Agent) so tasks never get lost in a backlog.
- Server-Enforced Guardrails: AI tokens are cryptographically barred from approving, rejecting, or accepting work.
- Cursor-Based Agent Inbox: Built for reliable polling. A crashed agent never loses an event or webhook notification.
- Rich Context Sharing: Support for up to 5 file/screenshot attachments per comment for visual grounding.
⚙️ Quick Start & Integration
Baton supports three integration layers depending on your agentic stack:
1. Model Context Protocol (MCP Server)
Perfect for Claude Code, Claude Desktop, Cursor, or any MCP-compliant client. Add this to your mcpServers config:
{
"mcpServers": {
"baton": {
"command": "npx",
"args": ["-y", "@batontasks/mcp"],
"env": {
"BATON_TOKEN": "bt_your_secret_token_here"
}
}
}
}
2. Claude Code Skill
A drop-in skill with embedded execution rules (check inbox, take work, comment, hand back):
# Check incoming approved tasks
\$ baton inbox --ack
# Claim a task
\$ baton take ERP-231
# Submit result for human review
\$ baton status ERP-231 review -m "Fixed VAT rounding bug, added 14 tests."
3. Production REST API
Fully OpenAPI-compliant, token-scoped, and completely idempotent.
curl https://batontasks.com \
-H "Authorization: Bearer bt_..."
📦 Board Migration
Moving from your old stack? Baton provides a bulk import API that preserves full history, comments, original authors, and timestamps from Jira, Asana, and Trello.
📄 License
Distributed under the MIT License. See LICENSE for more information.
Maintained by the Baton Team. Proudly built and tracked by a mixed human-agent workflow.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。