proc-repl-mcp
Enables stateful subprocess sessions (REPLs) and tmux-backed TUI program control for interactive tools like Python, r2, or shells.
README
proc-repl-mcp
Stateful subprocess sessions for MCP (Model Context Protocol).
中文文档: README.zh.md
This server is intentionally small: it keeps a subprocess alive across tool calls so you can interact with REPL-ish tools (python, r2/rizin, shells) and optionally drive full-screen TUI programs (vim/htop) via a tmux backend.
Security
This MCP server is a local RCE capability. By default nothing is allowed unless you set:
PROC_MCP_ALLOW: comma-separated allowlist of commands (e.g.python3,r2,sh)
Env overrides from clients are restricted by default (blocks PATH, LD_*,
DYLD_*, etc). See PROC_MCP_ENV_ALLOW / PROC_MCP_ENV_DENY in proc_repl_mcp.py.
Tools
open_session,list_sessions,send,read,run,close_sessiontmux_open_session,tmux_list_sessions,tmux_send_keys,tmux_step,tmux_capture_pane,tmux_resize,tmux_close_session
Cursor MCP config (uvx, no manual install)
If you have uv installed, you can run this without pre-installing the package:
{
"mcpServers": {
"proc-repl-mcp": {
"command": "uvx",
"args": ["-q", "-U", "proc-repl-mcp"],
"env": {
"PROC_MCP_ALLOW": "*"
}
}
}
}
Notes:
- For r2/rizin, use
-0/-q0andrun(until_nul=true)for reliable message boundaries. - For TUIs, prefer the tmux tools:
tmux_open_sessionthentmux_stepto send-keys and capture output in one roundtrip. tmux_send_keysnow sends each key as its own tmux operation and keeps a short gap beforeEnter. Tune it withPROC_MCP_TMUX_KEY_DELAY_MSif a TUI needs slower submission timing.PROC_MCP_ALLOW="*"enables everything (local RCE). Prefer a strict allowlist in real setups.
If uvx fails to connect to PyPI in your network, set a mirror index. You can
either pass it via args:
{
"mcpServers": {
"proc-repl-mcp": {
"command": "uvx",
"args": [
"-q",
"-U",
"--default-index",
"https://pypi.tuna.tsinghua.edu.cn/simple",
"proc-repl-mcp"
],
"env": {
"PROC_MCP_ALLOW": "*"
}
}
}
}
Or set it via env (recommended):
{
"mcpServers": {
"proc-repl-mcp": {
"command": "uvx",
"args": ["-q", "-U", "proc-repl-mcp"],
"env": {
"PROC_MCP_ALLOW": "*",
"UV_DEFAULT_INDEX": "https://pypi.tuna.tsinghua.edu.cn/simple"
}
}
}
}
If you haven't published to PyPI yet, you can run directly from GitHub (pin to a tag or commit):
{
"mcpServers": {
"proc-repl-mcp": {
"command": "uvx",
"args": [
"-q",
"-U",
"--from",
"git+https://github.com/SammySnake-d/proc-repl-mcp.git@main",
"proc-repl-mcp"
],
"env": {
"PROC_MCP_ALLOW": "*"
}
}
}
}
Install (pip)
python3 -m pip install proc-repl-mcp
proc-repl-mcp
Or run without installing (uvx):
uvx proc-repl-mcp
Dev
Run smoke tests locally:
python3 test_smoke.py
python3 test_full_smoke.py
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。