claude-bridge-mcp
An MCP server that exposes your local Claude Code CLI over HTTP+SSE, enabling any MCP-compatible client to use your Claude Code MAX/PRO subscription remotely.
README
claude-bridge-mcp
Use your Claude Code MAX/PRO subscription from anywhere.
An MCP server that exposes your local Claude Code CLI over HTTP+SSE, so any MCP-compatible client — OpenClaw, Claude Desktop, or your own agents — can use your Claude Code subscription remotely.
Your PC (Claude MAX/PRO) Remote Machine
┌──────────────────────┐ ┌──────────────────┐
│ claude-bridge-mcp │◄────────────►│ OpenClaw Agent │
│ :3100/sse │ HTTP+SSE │ Claude Desktop │
│ │ (Tailscale │ Custom MCP app │
│ Spawns Claude CLI ──┤ or VPN) └──────────────────┘
│ Uses YOUR sub ─────┤
└──────────────────────┘
Your $100/mo MAX or $200/mo PRO → accessible from any machine
Requirements
- Node.js 18+
- Claude Code CLI installed and authenticated (
claudecommand in PATH) — install guide - An active Claude Code MAX ($100/mo) or PRO ($200/mo) subscription
Quick Start
# Run directly (no install needed)
npx claude-bridge-mcp
# Or install globally
npm install -g claude-bridge-mcp
claude-bridge-mcp
The server starts on http://0.0.0.0:3100. Verify:
curl http://localhost:3100/health
Connect from OpenClaw
Add the bridge as an MCP server in your ~/.openclaw/openclaw.json:
{
"mcpServers": {
"claude-bridge": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://<your-pc-ip>:3100/sse"]
}
}
}
Replace <your-pc-ip> with your PC's IP address. If using Tailscale, use your Tailscale IP (tailscale ip -4).
Now your OpenClaw agent can use claude_execute, claude_query, claude_read_file, and claude_git_status — all powered by your Claude Code subscription.
Connect from Claude Desktop
Add to your Claude Desktop MCP config (claude_desktop_config.json):
{
"mcpServers": {
"claude-bridge": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://<your-pc-ip>:3100/sse"]
}
}
}
Available Tools
| Tool | Description |
|---|---|
claude_execute |
Execute a task using Claude Code. Full coding capabilities: read/write files, run commands, git operations. |
claude_query |
Ask Claude Code a question (read-only, fast). Great for code analysis and explanations. |
claude_read_file |
Read a file from the PC filesystem. |
claude_git_status |
Get git status: branch, changes, and recent commits. |
Configuration
All configuration is via environment variables. Copy .env.example to .env:
cp .env.example .env
| Variable | Default | Description |
|---|---|---|
BRIDGE_HOST |
0.0.0.0 |
Bind address |
BRIDGE_PORT |
3100 |
Server port |
BRIDGE_API_TOKEN |
(none) | Bearer token for authentication. If set, all requests (except /health) require Authorization: Bearer <token> |
BRIDGE_ALLOWED_IPS |
(empty = all) | Comma-separated IP allowlist. Empty means all IPs allowed. Localhost is always allowed. |
BRIDGE_ALLOWED_DIRS |
(empty = cwd) | Comma-separated directory allowlist. Empty defaults to the current working directory. |
BRIDGE_TIMEOUT |
120000 |
Execution timeout in ms |
BRIDGE_MAX_CONCURRENT |
2 |
Max concurrent Claude CLI executions |
BRIDGE_QUEUE_TIMEOUT |
30000 |
Queue wait timeout in ms |
Security
The bridge includes multiple security layers:
- IP Allowlist — Restrict access to specific IPs (e.g., your Tailscale network). Localhost always allowed.
- Bearer Token Auth — Set
BRIDGE_API_TOKENfor token-based authentication. - Directory Allowlist — Claude CLI can only access directories you explicitly allow.
- Symlink Protection — Paths are resolved via
realpath()before checking the allowlist, preventing symlink traversal. - Execution Queue — FIFO queue with configurable concurrency limits and timeout to prevent resource exhaustion.
Recommended setup for remote access:
# Use Tailscale for encrypted networking
BRIDGE_ALLOWED_IPS=100.x.y.z # Your remote machine's Tailscale IP
BRIDGE_API_TOKEN=your-secret # Additional auth layer
BRIDGE_ALLOWED_DIRS=/path/to/project1,/path/to/project2
Endpoints
| Endpoint | Method | Description |
|---|---|---|
/sse |
GET | SSE connection for MCP clients |
/messages |
POST | JSON-RPC message endpoint |
/health |
GET | Health check (no auth required) |
/metrics |
GET | Server metrics |
How It Works
- Remote MCP client connects to
/sse(Server-Sent Events) - Client sends tool calls via
/messages(JSON-RPC over HTTP) - Bridge spawns
claudeCLI locally with your authenticated session - Results stream back over SSE
The bridge uses the Model Context Protocol — the open standard for AI tool communication.
Supported Platforms
Runs anywhere Claude Code CLI and Node.js are available:
| Platform | Notes |
|---|---|
| Windows 10/11 | Full support |
| macOS (Intel & Apple Silicon) | Full support |
| Linux (Ubuntu, Debian, etc.) | Full support |
| Linux VPS (AWS, DigitalOcean, etc.) | Run the bridge on any cloud VM |
| Docker | node:18-alpine or similar |
| Mac Mini (headless server) | Great as an always-on bridge |
| Termux (Android) | Set BRIDGE_ALLOWED_DIRS explicitly |
| WSL2 | Full support |
Same-machine use: The bridge also works locally — useful for apps that only speak MCP but need Claude Code capabilities. Just connect to http://localhost:3100/sse.
Development
git clone https://github.com/0motionguy/claude-bridge-mcp.git
cd claude-bridge-mcp
npm install
npm run dev # Hot-reload development server
npm run build # Production build
License
MIT - see LICENSE
Built by ICM Motion for the OpenClaw community.
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。