Shift MCP Server
A lightweight coordination layer for multiple AI agents working on the same codebase, providing check-in and check-out tools via STDIO or Streamable HTTP.
README
<div align="center"> <h1>@cyanheads/shift-mcp-server</h1> <p><b>Lightweight coordination layer for multiple AI agents working on the same codebase. STDIO & Streamable HTTP</b></p> <p><b>2 Tools · 1 Resource</b></p> </div>
<div align="center">
</div>
Tools
Two tools for multi-agent coordination:
| Tool Name | Description |
|---|---|
shift_check_in |
Register or update a worker session. Returns worker ID, coordination instructions, and active peers. |
shift_check_out |
End a working session. Removes from active worker list. |
shift_check_in
Register a new worker or update an existing session. Called at the start of every working session.
- Accepts a gist of current work and optional file paths being modified
- Returns a 6-character worker ID, coordination protocol, and the full active workers table
- Pass an existing worker ID to update your session (patch semantics — omitted fields preserved)
- Error responses include the active workers table so agents can self-identify or start fresh
shift_check_out
End a working session and remove from the active worker list.
- Accepts a worker ID and optional summary of what was accomplished
- Idempotent — succeeds silently if the worker ID doesn't exist or was already checked out
Resources
| URI | Description |
|---|---|
shift://status |
All currently active workers with gists, declared files, and timestamps. |
Features
Built on @cyanheads/mcp-ts-core:
- Declarative tool definitions — single file per tool, framework handles registration and validation
- Unified error handling across all tools
- Structured logging with request-scoped context
- Runs locally (stdio/HTTP) from the same codebase
Coordination-specific:
- In-memory worker session store — no database, no filesystem writes, clears on restart
- Coordination protocol injected on every check-in so agents know how to behave
- Active workers table returned with every response for situational awareness
- Patch semantics on session updates — only provided fields change
Getting Started
MCP Client Config
Add to your MCP client config (e.g., claude_desktop_config.json):
{
"mcpServers": {
"shift": {
"type": "stdio",
"command": "bunx",
"args": ["@cyanheads/shift-mcp-server@latest"],
"env": {
"MCP_TRANSPORT_TYPE": "stdio"
}
}
}
}
Prerequisites
- Bun v1.2.0 or higher
Installation
git clone https://github.com/cyanheads/shift-mcp-server.git
cd shift-mcp-server
bun install
Configuration
No server-specific environment variables required. Framework defaults:
| Variable | Description | Default |
|---|---|---|
MCP_TRANSPORT_TYPE |
Transport: stdio or http. |
stdio |
MCP_HTTP_PORT |
Port for HTTP server. | 3010 |
MCP_HTTP_HOST |
Hostname for HTTP server. | 127.0.0.1 |
MCP_LOG_LEVEL |
Log level (RFC 5424). | info |
Running the Server
Local Development
bun run build
bun run start:stdio # or start:http
Dev mode with watch:
bun run dev:stdio # or dev:http
Checks and tests:
bun run devcheck # Lints, formats, type-checks
bun test # Runs test suite
Docker
docker build -t shift-mcp-server .
docker run -p 3010:3010 shift-mcp-server
Project Structure
| Directory | Purpose |
|---|---|
src/index.ts |
Entry point — registers tools and resources with createApp(). |
src/mcp-server/tools/definitions/ |
Tool definitions (check-in.tool.ts, check-out.tool.ts). |
src/mcp-server/tools/definitions/worker-store.ts |
In-memory worker session store and formatting utilities. |
src/mcp-server/resources/definitions/ |
Resource definitions (status.resource.ts). |
Development Guide
See CLAUDE.md for development guidelines and architectural rules. The short version:
- Handlers throw, framework catches — no
try/catchin tool logic - Use
ctx.logfor request-scoped logging - Register new tools and resources in
src/index.ts
Contributing
Issues and pull requests are welcome. Run checks before submitting:
bun run devcheck
bun test
License
Apache-2.0 — see LICENSE for details.
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。