Tweek MCP Server
An MCP server that integrates with the Tweek API to manage calendars, tasks, and custom colors. It enables users to perform full task CRUD operations, list accessible calendars, and fetch color preferences through secure authentication.
README
Tweek MCP Server
A Model Context Protocol (MCP) server that integrates with the Tweek API to manage calendars, tasks, and custom colors. Built with TypeScript and ESM modules.
Features
- Secure Authentication: Interactive sign-in with automatic token refresh
- Calendar Management: List calendars with role-based access
- Task CRUD: Create, read, update, and delete tasks with validation
- Custom Colors: Fetch user-specific color preferences
- Pagination Support: Forward-only pagination with
nextDocId - Error Handling: Structured error mapping with retry logic for transient failures
- Token Security: Encrypted token storage with file permissions (mode 600)
Prerequisites
- Node 20+
- PNPM 10+
- Tweek API key (obtain from Tweek)
Installation
# Install dependencies
pnpm install
# Build the project
pnpm build
Quick Start
1. Configure Environment
Set your Tweek API key (required):
export TWEEK_API_KEY="your-api-key-here"
See Configuration for all available environment variables.
2. First-Time Authentication
Before starting the server, you need to authenticate once:
pnpm auth:signin
You'll be prompted to enter your Tweek email and password:
Enter your Tweek email: user@example.com
Enter your Tweek password: ********
✅ Authentication successful! Tokens saved.
Your credentials are used only in memory and never persisted. The command exchanges them for tokens and securely saves them to ~/.tweek-mcp/tokens.json (or your configured TWEEK_TOKENS_PATH).
Alternative Authentication Methods
Non-interactive (for automation/CI):
echo "your-password" | pnpm auth:signin --email user@example.com --password-stdin
Import existing refresh token:
pnpm auth:import --refresh-token "your-refresh-token"
3. Start the Server
pnpm start
The server will:
- Load tokens from the token file
- Automatically refresh the
idTokenif needed - Start listening for MCP requests
For development with auto-reload:
pnpm dev
Configuration
Configure via environment variables:
| Variable | Required | Default | Description |
|---|---|---|---|
TWEEK_API_KEY |
✅ Yes | - | Your Tweek API key |
TWEEK_API_BASE |
No | https://tweek.so/api/v1 |
Tweek API base URL |
TWEEK_TOKENS_PATH |
No | ~/.tweek-mcp/tokens.json |
Path to token storage file |
TWEEK_REQUEST_TIMEOUT_MS |
No | 30000 |
HTTP request timeout (milliseconds) |
TWEEK_TOKEN_REFRESH_BUFFER_SEC |
No | 60 |
Proactive token refresh window (seconds) |
TWEEK_ENCRYPTION_KEY |
No | - | 32-byte key for AES-GCM token encryption |
Example Configuration
export TWEEK_API_KEY="ak_live_..."
export TWEEK_TOKENS_PATH="/secure/volume/tweek-tokens.json"
export TWEEK_ENCRYPTION_KEY="your-32-byte-encryption-key-here"
export TWEEK_REQUEST_TIMEOUT_MS="45000"
export TWEEK_TOKEN_REFRESH_BUFFER_SEC="120"
MCP Tools
The server exposes the following MCP tools:
Calendars
listCalendars()
List all calendars accessible to the authenticated user.
Returns:
interface ListCalendarsResponse {
calendars: Calendar[]
}
interface Calendar {
id: string
name: string
role: 'ROLE_OWNER' | 'ROLE_EDITOR' | 'ROLE_VIEWER'
color?: string
// ... other calendar fields
}
Tasks
listTasks({ calendarId, startAt?, dateFrom?, dateTo? })
List tasks with optional pagination and date filtering.
Parameters:
calendarId(required): Calendar IDstartAt(optional): Pagination cursor from previous responsedateFrom(optional): ISO 8601 date to filter tasks fromdateTo(optional): ISO 8601 date to filter tasks to
Returns:
interface ListTasksResponse {
pageSize: number
nextDocId?: string
data: Task[]
}
getTask({ taskId })
Retrieve a single task by ID.
createTask({ task })
Create a new task.
Parameters:
task.calendarId(required): Calendar IDtask.text(required): Task titletask.date(optional): ISO 8601 datetask.freq(optional): Recurrence frequency (0-7)task.checklist(optional): Array of checklist items- ... see Tweek API docs for all fields
Returns:
interface CreateTaskResponse {
id: string
}
updateTask({ taskId, patch })
Update an existing task.
Returns: Full updated task object
deleteTask({ taskId })
Delete a task.
Returns:
interface DeleteTaskResponse {
success: true
}
Custom Colors
getCustomColors()
Fetch custom colors for the authenticated user. The userId is automatically extracted from the idToken.
Returns:
interface GetCustomColorsResponse {
colors: Color[]
}
interface Color {
id: string
hex: string
name?: string
}
Input Validation
The server validates task inputs:
freq: Must be integer 0-7 (Tweek recurrence enum)notifyAt,date,isoDate,dtStart: Must be valid ISO 8601 date/datetimecalendarId: Required non-empty string on creationchecklist: Each item must have non-emptytextfield
Connecting MCP Clients
Claude Desktop
Add to your Claude Desktop configuration (claude_desktop_config.json):
{
"mcpServers": {
"tweek": {
"command": "node",
"args": ["/path/to/tweek-mcp/dist/index.js"],
"env": {
"TWEEK_API_KEY": "your-api-key-here"
}
}
}
}
Other MCP Clients
The server uses standard MCP JSON-RPC over stdio. Configure your client to:
- Run
node dist/index.js(orpnpm start) - Pass
TWEEK_API_KEYin environment - Ensure token file exists at configured path
Development
# Install dependencies
pnpm install
# Run tests
pnpm test
# Run linter
pnpm lint
# Build TypeScript
pnpm build
# Development mode (auto-reload)
pnpm dev
Project Structure
src/
├── auth/ # Authentication (AuthManager, TokenStore, IdentityClient)
├── cli/ # CLI commands (auth signin/import)
├── config/ # Configuration loading
├── http/ # HTTP client with retry logic
├── logging/ # Structured logging
├── tests/ # Test suite
├── tools/ # MCP tools (calendars, tasks, colors, validation)
├── tweek/ # Tweek API client and types
└── index.ts # Server bootstrap
Troubleshooting
Error: "Tweek tokens not found"
Problem: Server cannot find the token file.
Solution: Run authentication first:
pnpm auth:signin
Error: "UNAUTHENTICATED"
Problem: Tokens are expired or invalid.
Solution: Re-authenticate:
pnpm auth:signin
Error: "Permission denied" on token file
Problem: Token file permissions are incorrect.
Solution: The server automatically sets mode 600. If issues persist, manually fix:
chmod 600 ~/.tweek-mcp/tokens.json
Tokens Not Refreshing
Problem: Token refresh fails silently.
Solution: Check logs for refresh errors. Verify your refresh token is still valid by re-authenticating.
Connection Timeouts
Problem: Requests timing out to Tweek API.
Solution: Increase timeout:
export TWEEK_REQUEST_TIMEOUT_MS="60000"
Container Deployments
For Docker or containerized environments:
-
Mount a persistent volume for tokens:
docker run -v /secure/volume:/tokens \ -e TWEEK_TOKENS_PATH=/tokens/tweek.json \ -e TWEEK_API_KEY=your-key \ tweek-mcp -
Consider using
TWEEK_ENCRYPTION_KEYfor additional security -
Pre-authenticate using
auth:importwith a refresh token
Security
- Tokens stored with file permissions 600 (user-only access)
- Optional AES-GCM encryption for token file
- Credentials never logged or persisted
- Sensitive headers redacted from logs
- Automatic token refresh before expiry
Testing
Comprehensive test suite with:
- Unit tests for validation, retry logic, encryption
- Integration tests for HTTP client and auth flows
- CLI end-to-end tests
- Contract tests for MCP tools
# Run all tests
pnpm test
# Run specific test file
pnpm test src/tests/validation.test.ts
# Run tests matching pattern
pnpm test -- -t "should validate task input"
Contributing
- Follow the existing code style (enforced by ESLint)
- Add tests for new features
- Ensure
pnpm lintandpnpm testpass - Update documentation as needed
License
MIT
Resources
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。