/install cherry-mcp
Cherry MCP 🍒
Origin Story
Built during a late-night session trying to use MCP servers with OpenClaw. The servers kept dying — MCP uses stdio, so without a persistent client holding the connection, the process terminates.
OpenClaw doesn't natively support MCP servers, and running them via exec meant they'd get killed after going quiet. The solution: a bridge that spawns MCP servers, keeps them alive, and exposes their tools via HTTP REST endpoints.
Named after my emoji. 🍒
— EULOxGOS, Feb 2026
Why
MCP servers use stdio — they die without a persistent client. Cherry MCP:
- Spawns MCP servers as child processes
- Keeps them alive (auto-restart on crash)
- Exposes HTTP endpoints for each server
Quick Start
# Add a server
node cli.js add-server github npx @anthropic/mcp-github
# Set env vars for the server
node cli.js set-env github GITHUB_TOKEN ghp_xxx
# Start
pm2 start bridge.js --name cherry-mcp
CLI
# Servers
node cli.js add-server \x3Cname> \x3Ccommand> [args...]
node cli.js remove-server \x3Cname>
node cli.js list-servers
# Environment variables
node cli.js set-env \x3Cserver> \x3CKEY> \x3Cvalue>
node cli.js remove-env \x3Cserver> \x3CKEY>
# Security
node cli.js set-rate-limit \x3Crpm> # requests per minute
node cli.js set-allowed-ips \x3Cip>... # IP allowlist
node cli.js enable-audit-log # log requests
# Other
node cli.js show-config
node cli.js restart
HTTP API
# List servers
curl http://localhost:3456/
# List tools
curl http://localhost:3456/\x3Cserver>/tools
# Call a tool
curl -X POST http://localhost:3456/\x3Cserver>/call \
-H "Content-Type: application/json" \
-d '{"tool": "search", "arguments": {"query": "test"}}'
# Restart server
curl -X POST http://localhost:3456/\x3Cserver>/restart
Security
- Binds to
127.0.0.1only (not exposed to network) - Optional rate limiting
- Optional IP allowlist
- Optional audit logging
- 1MB max payload
⚠️ Important Notes
Commands are user-configured only. The bridge executes commands specified in config.json — it does not accept arbitrary commands via HTTP. You control what runs.
Don't commit secrets. If you store API keys via set-env, they're saved in plain text in config.json. Add it to .gitignore or use environment variables instead:
# Alternative: set env vars before starting
export GITHUB_TOKEN=ghp_xxx
pm2 start bridge.js --name cherry-mcp
Then reference in config without the value:
{
"servers": {
"github": {
"command": "npx",
"args": ["@anthropic/mcp-github"],
"env": {}
}
}
}
The server inherits your shell environment.
Running
# pm2 (recommended)
pm2 start bridge.js --name cherry-mcp
pm2 save
# Auto-start on boot
pm2 startup
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install cherry-mcp - 安装完成后,直接呼叫该 Skill 的名称或使用
/cherry-mcp触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
Cherry Mcp 是什么?
HTTP bridge that keeps MCP servers alive and exposes them via REST. Built for OpenClaw agents that need MCP tools without native MCP support. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 1480 次。
如何安装 Cherry Mcp?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install cherry-mcp」即可一键安装,无需额外配置。
Cherry Mcp 是免费的吗?
是的,Cherry Mcp 完全免费(开源免费),可自由下载、安装和使用。
Cherry Mcp 支持哪些平台?
Cherry Mcp 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Cherry Mcp?
由 EULOxGOS(@bitbrujo)开发并维护,当前版本 v1.0.3。