TaskMCP
A personal task management MCP server that allows LLM clients to create, read, update, and delete tasks with projects, labels, and comments, using a local SQLite database that can also be accessed via a web UI.
README
TaskMCP
A personal task management system with a lightweight Web UI and a Model Context Protocol (MCP) server.
Features
- Local-first SQLite storage.
- MCP server for LLM clients (stdio transport).
- Terminal-inspired Web UI with React + Tailwind CSS v4.
- Drag-and-drop task cards across status columns.
- Inline card editing and per-column task creation.
- Projects, tasks, labels, and comments.
- Both LLM and browser write to the same database.
Architecture
TaskMCP runs as two independent processes that share a single SQLite file:
npm run mcp— MCP server on stdio for LLM clients.npm run web— Hono web server onhttp://localhost:3000.
Both processes open the same better-sqlite3 database configured with WAL mode and a 5 second busy timeout, so they can run concurrently without blocking each other.
Installation
npm install
Scripts
npm run build— Compile TypeScript.npm run typecheck— Run TypeScript without emitting files.npm run test— Run Vitest tests.npm run mcp— Start the MCP server (stdio).npm run web:build— Build the React client and Tailwind CSS output.npm run web:watch— Run the web server with automatic client rebuilds on file changes.npm run web:serve— Start only the Hono API/static server.npm run web— Build and start the web UI onhttp://localhost:3000.npm run css:build— Build the Tailwind CSS output.npm run design:lint— ValidateDESIGN.md.npm run design:export— Export design tokens tosrc/web/static/theme.css.npm run lint— Run ESLint.npm run format— Format files with Prettier.
Design System
Visual tokens are defined in DESIGN.md. Export them and rebuild CSS with:
npm run design:export
npm run css:build
The front-end is built with React, bundled by esbuild, and styled with Tailwind CSS v4 classes generated from src/web/client/index.css.
Note: The
@google/design.mdv0.3.0exportCLI does not emit CSS correctly, sodesign:exportruns a small local script (scripts/generate-theme.js) that parsesDESIGN.mdand generates the equivalent Tailwind v4@themeblock.
Configuration
Set TASKMCP_DB_PATH to override the default SQLite database location (.data/taskmcp.db).
API Endpoints
All endpoints are mounted under /api:
| Method | Endpoint | Description |
|---|---|---|
| GET | /api/projects |
List all projects |
| GET | /api/labels |
List all labels |
| GET | /api/tasks |
List tasks (query: projectId, status, label, query) |
| GET | /api/tasks/:id |
Get a single task |
| POST | /api/tasks |
Create a task |
| PATCH | /api/tasks/:id |
Update a task |
| DELETE | /api/tasks/:id |
Delete a task |
| POST | /api/tasks/:id/comments |
Add a comment |
Usage Workflows
Add a task through your LLM client
- Configure the MCP client (see below).
- Ask it to create a task:
在收件箱项目里创建一个标题为 "Review design system" 的高优先级任务。 - The task is written to
.data/taskmcp.db.
View and edit tasks in the browser
- Start the web server:
npm run web - Open
http://localhost:3000. - Use the search box, status filter, or label filter to find tasks.
- Click a task card to edit, change status, add comments, or delete it.
Both interfaces share the same data
Because MCP and Web open the same SQLite file, any change from one side appears in the other immediately after a refresh.
MCP Clients
Replace /path/to/CcTaskMCP with the actual project path.
Claude Desktop (claude_desktop_config.json)
{
"mcpServers": {
"taskmcp": {
"command": "npx",
"args": ["tsx", "/path/to/CcTaskMCP/src/mcp/index.ts"]
}
}
}
opencode (opencode.json)
{
"mcp": {
"taskmcp": {
"command": "npx",
"args": ["tsx", "/path/to/CcTaskMCP/src/mcp/index.ts"]
}
}
}
Note on the MCP SDK
The plan originally referenced @modelcontextprotocol/sdk@^2.x; npm currently ships 1.29.0 as the latest stable release. The server is implemented with the v1 McpServer API and is functionally equivalent for this project.
License
MIT
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。