claude-mux.mcp
Gives Claude Code terminal control and multi-agent coordination through tmux sessions.
README
claude-mux.mcp
MCP server for tmux. Gives Claude Code terminal control and multi-agent coordination through tmux sessions.
Install
claude plugin add claude-mux@rtd
Then run /install inside Claude Code to verify dependencies and register the MCP server. It checks for bun and tmux, installs packages, and smoke-tests the server.
To update:
claude plugin update claude-mux@rtd
Then run /install again to verify everything is current.
Requirements
- tmux —
brew install tmuxon macOS,apt install tmuxon Linux - Bun —
curl -fsSL https://bun.sh/install | bash
Manual install (development)
If you're working on the server itself, add to ~/.claude.json:
{
"mcpServers": {
"tmux": {
"command": "bun",
"args": ["/path/to/claude-mux.mcp/server.js"]
}
}
}
Actions
Observe
| Action | What it does |
|---|---|
list |
All sessions. Named ones show windows; numbered ones get a one-line summary. |
session |
One session in detail, previewing each pane's last visible line. |
read |
Capture pane output. lines sets history depth (default 100). |
tail |
Last N lines, no pagination. Quick look at what just happened. |
watch |
Delta since last read. Only new lines come back. |
transcript |
Read a pane's prior Claude Code sessions from the on-disk JSONL — not scrollback. No name: lists the cwd's sessions newest-first with a title preview. With name (list index or session id): the flattened conversation, tool calls reduced to one-line breadcrumbs. |
layout |
Every pane across all sessions with dimensions and running process. |
meta |
Authoritative machine-parseable metadata for dashboards (no params). Tab-delimited rows, row-type letter in column 0 — one S row per session (named and numbered, no collapsing) and one W row per window. S\t<session_name>\t<session_activity>\t<session_attached>\t<session_created>; W\t<session_name>\t<window_index>\t<window_name>\t<window_activity>\t<window_active 0|1>. Epochs are raw seconds; session_attached is the raw client count (>0 = attached). TAB is the sole delimiter (window names may contain spaces and :). |
Act
| Action | What it does |
|---|---|
type |
Literal text, spaces preserved. No Enter appended. |
typewait |
Type literal text + Enter, then wait. One call for "run this and show me what happened." |
keys |
Space-separated key tokens: Escape :q! Enter, C-c, Down Down Enter. |
keyswait |
Send keys, poll until prompt appears or output settles. Returns the delta. |
exec |
Wrap a command in start/done markers, return a commandId. Non-blocking. |
result |
Check a tracked command's status, exit code, and output by commandId. |
Manage
| Action | What it does |
|---|---|
new-session |
Create a session. |
kill-session |
Kill a session. |
new-window |
Create a window in a session. Returns the target for subsequent calls. |
kill-window |
Kill a window. |
split |
Split a pane. Returns the new pane target. text: "horizontal" or "vertical" (default). |
kill-pane |
Kill a pane. |
rename |
Rename a window. |
Coordinate
Cross-session agent messaging and task ownership. All state lives in tmux global environment and named buffers.
| Action | What it does |
|---|---|
register |
Identify yourself by name. Stored globally so other agents can find you. |
unregister |
Remove your registration. |
who |
List all registered agents across all sessions. |
post |
Message an agent by name, or "all" for broadcast. |
inbox |
Check for new messages. |
claim |
Take a task. Atomic file lock prevents two agents from claiming the same one. |
complete |
Mark a task done, release the lock. |
tasks |
List all tasks with owner and status. |
Workers
Spawn Claude Code instances as coordinating agents. Each worker gets a name, a team roster, and auto-injected inbox hooks so messages arrive without polling.
| Action | What it does |
|---|---|
spawn |
One-shot claude -p. Auto-closes when done, captures output to a tmux buffer. |
spawn-persist |
Same, but the window stays open for inspection. |
teammate |
Interactive claude session. Stays alive for ongoing coordination. |
despawn |
Kill a worker, clean up registration and temp files. |
worker-result |
Read a completed worker's captured output. |
Design notes
Plain text responses. list returns indented text, not nested JSON. Fewer tokens, easier for the model to parse.
Self-awareness. The server reads $TMUX_PANE on startup and reports you are here: main:0.1 in listings so the model knows which pane is itself.
type vs keys. keys splits on spaces — each token is a tmux key name. Good for Escape :q! Enter, destroys prose. type sends literal text with spaces intact.
exec/result for long commands. typewait blocks up to 30 seconds. exec drops start/done markers, returns a commandId immediately, and result checks on it later. Auto-detects zsh, bash, or fish.
Completion detection. keyswait and typewait poll every 500ms. Shell prompt means done. Output unchanged for 2s means settled. 30s timeout ceiling.
Atomic task claiming. claim uses O_EXCL file creation so two agents racing for the same task can't both win.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.