Terminal MCP Server
Enables AI agents to spawn, interact with, and orchestrate multiple concurrent terminal sessions via tmux.
README
Terminal MCP Server
MCP server for managing multiplexed terminal sessions via tmux. Lets AI agents spawn, send input to, read output from, and orchestrate multiple concurrent terminal sessions.
Why
Claude Code (and other MCP clients) can run shell commands, but can't manage long-running processes, interactive sessions, or multiple concurrent terminals. This server bridges that gap using tmux as the multiplexer.
Key use case: multi-agent orchestration. Run multiple Claude Code sessions in tmux panes, then use terminal_send to type messages into any session — from the target's perspective, a human just typed. This enables Claude-to-Claude communication with full context and permissions.
Install
Requires tmux installed and in PATH.
Add to your Claude Code MCP config (~/.claude/mcp.json or .mcp.json):
{
"mcpServers": {
"terminal": {
"command": "npx",
"args": ["@quantum-encoding-europe-limited/terminal-mcp"]
}
}
}
Or run directly:
npx @quantum-encoding-europe-limited/terminal-mcp
Tools
| Tool | Description |
|---|---|
terminal_list |
List all active sessions (name, PID, cwd, command) |
terminal_spawn |
Create a named session with optional cwd, startup command, and mode |
terminal_send |
Type text into a session (keystrokes via tmux send-keys) |
terminal_read |
Read last N lines from a session (ANSI-stripped) |
terminal_close |
Graceful close (Ctrl+C, Ctrl+D, then kill) |
terminal_send_signal |
Send Ctrl+C/D/Z/L/\ to a session |
terminal_resize |
Resize pane dimensions |
terminal_snapshot |
Overview of ALL sessions — last 5 lines each |
terminal_wait |
Poll until output matches a pattern (e.g. "Server ready") |
Spawn Modes
terminal_spawn supports a mode parameter for controlling how Claude sessions run:
| Mode | Behaviour |
|---|---|
interactive |
(default) Normal foreground session with permission prompts. Best when you need to approve tool calls. |
background |
Auto-allows safe tools (Bash, Edit, Read, Write, Glob, Grep, Agent) but blocks destructive ops (git force push, git reset --hard, git branch -D, git clean -f, rm -rf). Best for autonomous work with guardrails. |
resume |
Resume an existing Claude session by name, with the same background safety guardrails. |
terminal_spawn("scanner", cwd="/my/project", mode="background")
// Launches: claude --allowedTools "Bash(*) Edit Read Write Glob Grep Agent"
// --disallowedTools "Bash(git push --force*) Bash(git reset --hard*) ..."
terminal_spawn("scanner", mode="resume")
// Launches: claude --resume "scanner" --allowedTools ... --disallowedTools ...
Note: Background mode never uses --dangerously-skip-permissions. It uses curated allow/deny lists so agents can work autonomously without being able to force-push, delete branches, or wipe files.
Examples
Run a dev server and wait for it
terminal_spawn("server", cwd="/my/project", command="npm run dev")
terminal_wait("server", pattern="ready on port 3000")
Orchestrate multiple Claude sessions
# Spawn specialists in background mode (no permission prompts)
terminal_spawn("backend", cwd="/work/api", mode="background")
terminal_spawn("frontend", cwd="/work/app", mode="background")
# Wait for Claude to start, then send tasks
terminal_wait("backend", pattern="? for shortcuts")
terminal_send("backend", "add input validation to the POST /users endpoint")
terminal_wait("frontend", pattern="? for shortcuts")
terminal_send("frontend", "update the user form to match the new API validation")
Monitor everything at once
terminal_snapshot() # see last 5 lines from every session
Configuration
Environment variables:
| Variable | Default | Description |
|---|---|---|
TERMINAL_MCP_SESSION |
auto-detected / mcp-terminals |
tmux session name |
TERMINAL_MCP_MAX_LINES |
200 |
Default line capture limit |
If running inside tmux, the current session is auto-detected. Otherwise falls back to TERMINAL_MCP_SESSION or creates mcp-terminals.
How it works
All terminal management goes through tmux:
terminal_spawn→tmux split-window+tmux select-pane -Tterminal_send→tmux send-keys(short input) ortmux load-buffer+paste-buffer(long input, avoids escaping issues)terminal_read→tmux capture-pane -pterminal_close→tmux kill-pane
Name matching is fuzzy — handles tmux title prefixes (like Claude's spinner characters ✳ ✽ ⏺) automatically.
License
MIT
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。