TaskPilot MCP Server
Provides 13 tools for AI agents to manage tasks, calendar, and goals in TaskPilot, a local-first desktop task manager with a permissioned agent runner and review workflows.
README
TaskPilot
A local-first desktop task & calendar manager built to treat an AI agent as a first-class operator — not a chat bolted onto a to-do list.
TaskPilot runs entirely on your machine. Your tasks, calendar, and history live in a local SQLite database; an AI agent can read and act on them through a real API and a permissions model — the same way you can. It's what a task manager looks like when you design for a human and an agent working the same queue.
<!-- Add a screenshot at docs/screenshot.png and reference it here -->
Why it exists
Most "AI to-do apps" are a chat window that returns text. TaskPilot inverts that: the agent gets first-class, permissioned access to the same surfaces you use — create tasks, derive sub-tasks, execute agent-eligible work, manage your calendar — through four interfaces that all hit the same core:
- MCP server — 13 tools, so any MCP-aware client (Claude, etc.) can operate TaskPilot directly.
- HTTP API — a local REST API on
:7847with bearer auth. - File I/O protocol — a watched
io/directory, so agents can work even when the app is closed. - Desktop GUI — an Electron app with a keyboard-first, review-gated workflow.
Features
- Local-first & private — SQLite on disk, no cloud dependency, no account required.
- Agent runner with a permissions model — every agent action is scoped by an explicit permission profile (allowed tools, workspace, network, denied globs). Nothing runs unsandboxed.
- Derive → review → execute loop — break a task into sub-tasks, approve them, and let an agent execute the agent-eligible ones. Irreversible actions are gated behind human review.
- Calendar + availability — events, free/busy, and scheduling that tasks can hook into.
- Goals & projects — group work into initiatives and track progress.
- Integrations — GitHub, Gmail (draft-only by default), Slack, Notion (one-way mirror), calendar.
- Full-text search, cost tracking, decision logs, phase pipelines for content/dev work.
Stack
TypeScript (strict) · Electron · React + Tailwind · better-sqlite3 + Drizzle · Model Context Protocol SDK · Vite · Vitest
Getting started
npm install
npm run dev # Vite + Electron in dev
# or build and run:
npm run build && ELECTRON_PROD=1 npx electron .
The HTTP API comes up on http://localhost:7847; MCP server via npm run mcp.
Architecture
┌─────────────┐ ┌──────────────┐ ┌─────────────┐ ┌──────────────┐
│ Electron │ │ HTTP API │ │ MCP server │ │ File I/O │
│ GUI │ │ :7847 │ │ (13 tools) │ │ io/ watcher │
└──────┬──────┘ └──────┬───────┘ └──────┬──────┘ └──────┬───────┘
└─────────────────┴──────────────────┴─────────────────┘
│
┌────────────▼────────────┐
│ Core: tasks, calendar, │
│ goals, permissions, │
│ agent runner (SQLite) │
└──────────────────────────┘
License
MIT
Built by Jack Gierlich. TaskPilot is a working tool I use daily; it's shared as an example of local-first, agent-operable software design.
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。