pi-delegate-mcp
Delegates bounded coding tasks from MCP clients to the Pi Coding Agent over stdio. Supports review, verification, implementation, and batch operations with long-running task polling.
README
pi-delegate-mcp
Delegate bounded coding tasks from MCP clients (Cursor) to Pi Coding Agent over stdio via the official SDK (@earendil-works/pi-coding-agent).
Requirements
- Node.js 22.19+
- Existing
~/.pi/agent/auth.json(Codex OAuth) is reused — no separate Pi binary is required
Install
npm install -g pi-delegate-mcp
pi-delegate-mcp install cursor --scope global
pi-delegate-mcp doctor
pi-delegate-mcp auth status
Then restart Cursor. The server registers as pi-delegate in ~/.cursor/mcp.json.
Tools
| Tool | Purpose |
|---|---|
delegate_review |
Read-only review / static hunt (async; optional perspectives) |
delegate_verify |
Build / test / lint (async) |
delegate_implement |
Implement in worktree (async; default patch) |
delegate_judge |
Judge with no-tools profile (async) |
delegate_manual |
Manual prompt under fixed profile (async) |
delegate_batch |
Multi-task parallel/sequential batch |
delegate_roles |
Role-based pipeline (implement → verify → reviews) |
get_run / cancel_run |
Poll / cancel a single run |
get_batch / cancel_batch |
Poll / cancel a batch |
smoke_test |
SDK connectivity / OAuth / provider check (stdout must be OK) |
Long-running tools return { status: "running", runId|batchId } immediately. Poll with get_run / get_batch until complete (avoids Cursor MCP client timeouts).
Effort
med | high | xhigh | max → Pi thinking medium | high | xhigh | max.
Safety
- SDK tool allowlist is a model tool allowlist, not an OS sandbox. Policy extensions additionally block dangerous commands.
- Implement defaults to worktree + patch (does not modify your tree).
- Manual prompts cannot widen tools.
- Ambient skills / extensions / AGENTS.md are not loaded.
- No auto commit, push, PR, or deploy.
Config
pi-delegate-mcp config path prints the JSONC config location (version 2).
Auth:
pi-delegate-mcp auth status
pi-delegate-mcp auth login openai-codex
pi-delegate-mcp auth logout openai-codex
Development
bun install
bun run test
bun run build
bun run check:delegate-pi-assets
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。