io.github.pooriaarab/vibelive

Official registry Developer tools

Multiplayer for agentic coding terminals: live cursors, presence, chat.

What can you do with io.github.pooriaarab/vibelive?

  • Claude Code pairs: share process output and hand off the driver role on a trusted LAN.
  • Codex reviewers: observe a wrapped command while the host controls input.
  • Gemini CLI mentors: keep discussion in chat and send agent input through /type.
  • Node.js teams: wrap a line-based script to test collaboration without AI credentials.
  • Agent builders: start sessions through the Model Context Protocol (MCP) and inspect participants and driver state.

Install

Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).

NPM package vibelive
Claude Code
claude mcp add vibelive -- npx -y vibelive@0.2.2 mcp
JSON configuration (Claude Desktop, Cursor, VS Code)
{
    "mcpServers": {
        "vibelive": {
            "command": "npx",
            "args": [
                "-y",
                "vibelive@0.2.2",
                "mcp"
            ]
        }
    }
}

Documentation

From the project's README on GitHub, refreshed weekly. Links open on GitHub.

Contents
  1. vibelive
  2. Install
  3. Share your first process
  4. How it works
  5. Why it exists
  6. Compare
  7. Coverage
  8. Common use cases
  9. Features
  10. Runtime surfaces
  11. Delivery and behavior limits
  12. Integration guides
  13. Status
  14. FAQ
  15. License

vibelive

Package · Setup · API · Compare

vibelive is an open-source terminal collaboration tool for developers who share agentic coding sessions. It wraps a process, shares output over local/LAN WebSockets, and lets one driver send input at a time. It ships as a CLI, Node.js library, and agent server in the Vibe Suite.

Install

npm install -g vibelive

Requires Node.js 18 or later. Includes TypeScript declarations and three direct runtime dependencies.

Share your first process

In one terminal, start a process that echoes each input line:

vibelive host --port 4474 --name ada -- node -e "process.stdin.on('data', b => process.stdout.write(b))"

In another terminal, join it:

vibelive join ws://localhost:4474 --name bee

Type hello in the host terminal. Both terminals print hello; the joiner also shows presence and control state. No credentials or environment variables are required. vibelive does not load .env files; wrapped agents inherit the host environment. For LAN peers, use the host's printed LAN address. Transport is plain WebSocket with no built-in authentication or encryption.

In the joiner, type /drive. On the host, type /release to hand over control. The joiner can then send /type ping and see ping in both terminals. Use /quit on the host to end the session.

How it works

Host terminal -> wrapped child process -> ordered output log
                       ^                         |
                       |                         v
                 one driver <- host WebSocket relay -> participants
                                  |                    |
                              control queue        chat + presence

The host owns process output and control state. Late joiners receive retained output followed by live updates. The relay binds to 0.0.0.0 by default, so reachable LAN clients can connect.

Why it exists

Pairing on a coding agent needs a clear answer to who can type next. vibelive adds a cooperative driver queue and chat around a wrapped command. Tools such as tmux and tmate serve broader terminal workflows; vibelive focuses on process sharing with agent-accessible session management.

Compare

Checked 2026-10-14 against tmux and tmate.

Propertyvibelivetmuxtmate
ScopeShared process with driver arbitrationTerminal multiplexerTerminal sharing
Transport herePlain local/LAN WebSocketLocal sessions; remote access arranged separatelySSH-based sharing
InterfaceCLI, Node.js library, agent toolsTerminal commandsTerminal commands
Process handlingChild-process pipesTerminal sessionstmux-derived terminal sessions

Coverage

Surface or integrationStatusScope
WebSocket / wsImplementedHost relay, client, output, chat, presence, control
Claude CodeGeneric command wrapperRequires separately installed claude; no dedicated adapter
OpenAI CodexGeneric command wrapperRequires separately installed CLI; no dedicated adapter
Google Gemini CLIGeneric command wrapperRequires separately installed CLI; no dedicated adapter
Pi / Grok workflowsGeneric command wrapperPipes-compatible commands; no dedicated adapter
KimiGeneric command wrapperRequires separately installed CLI; no dedicated adapter
MCPImplementedTwo stdio tools
Cursor updatesProtocol onlyLibrary send/receive; no rich CLI cursor renderer
Encrypted remote relayRoadmapNot shipped
Full terminal emulationRoadmapNo pseudoterminal support

There is no harness adapter registry: the host executes an arbitrary command through Node.js child-process pipes. Local tests cover host, relay, client, arbitration, protocol-facing behavior, and MCP. Specific agent CLIs are not certified here.

Common use cases

  1. Claude Code pairs: share process output and hand off the driver role on a trusted LAN.
  2. Codex reviewers: observe a wrapped command while the host controls input.
  3. Gemini CLI mentors: keep discussion in chat and send agent input through /type.
  4. Node.js teams: wrap a line-based script to test collaboration without AI credentials.
  5. Agent builders: start sessions through the Model Context Protocol (MCP) and inspect participants and driver state.

Features

  • Single driver: first-in, first-out control queue; non-driver input is rejected.
  • Output replay: sequence-numbered retained history plus live output.
  • Chat: separate participant messages and process input.
  • Presence: participant names and the current driver.
  • Consent: relay output sharing checks vibe-core's share:session scope.
  • Lifecycle: host exit closes sessions; MCP disconnect cleans up its hosted processes.

Runtime surfaces

vibelive host [--port N] [--name name] -- command [args...]
vibelive join ws://host:port [--name name]
vibelive mcp
vibelive --help / -h
vibelive --version / -v

All wrapped-command flags go after --. --port 0 selects an available port. The join parser also accepts --port, but does not use it; the URL controls the destination. Join names default to USER, then USERNAME, then anon.

Interactive commandHostJoiner
/driveRequest controlRequest control
/releaseRelease controlRelease control
/quitEnd sessionLeave session
/type textTreated as process input if driverSend process input if driver
/helpTreated as process input if driverPrint slash-command help
Other textProcess input if driverChat

vibelive mcp starts stdio MCP with exactly two tools:

ToolInputResult
host_sessionCommand array; optional name and portSession ID and WebSocket join URL
session_statusOptional URL filterActive sessions, participants, and driver
Library example

Install locally with npm install vibelive, then save this as demo.mjs and run node demo.mjs:

import { createHost, createRelay, joinSession } from 'vibelive';
const host = createHost({ command: ['node', '-e', "setInterval(() => console.log('hello'), 200)"] });
const relay = await createRelay({ host: '127.0.0.1', port: 0, hostHandle: host });
const client = joinSession({ url: relay.url, name: 'ada' });
client.onOutput(text => process.stdout.write(text));
await client.id;
await new Promise(resolve => setTimeout(resolve, 600));
client.close();
host.kill();
await relay.close();

The script prints hello and closes its process and relay.

Delivery and behavior limits

The shipped transport is local/LAN, plain WebSocket. There is no built-in participant authentication, TLS, or end-to-end encryption. Use it only where session access and network exposure match your intended participants. Output can include anything the wrapped command prints. The host uses pipes, not a pseudoterminal (PTY). Resize is a no-op; raw-mode and full-screen agent interfaces may not work. Output history defaults to 5,000 chunks in memory. Older entries are evicted; sessions do not persist after host exit. Cursor messages exist in the protocol, but the CLI renders presence rather than animated cursors. The package provides no managed relay, remote fan-out scale guarantee, or durable chat service.

Integration guides

  • Claude Code process sharing: vibelive host -- claude; install and authenticate Claude Code separately.
  • Other coding agents: pass the installed command after --; check that it can operate through pipes.
  • LAN participation: use the printed ws://<LAN-address>:<port> URL with vibelive join.
  • MCP host configuration:
{"mcpServers":{"vibelive":{"command":"vibelive","args":["mcp"]}}}

Status

Package version 0.2.3 is pre-1.0; interfaces can change. The CLI's source version still reports 0.2.2. Planned work includes PTY wrapping, richer cursors, and an encrypted self-hosted remote relay. Public case studies are not yet published. No public documentation site is configured; setup guidance is self-contained here. Use GitHub issues for support; repository access may be required. For contributions, run npm ci, npm run typecheck, npm run build, and npm test.

FAQ

Which integrations are verified?

vibelive tests its local process and WebSocket behavior. Named coding agents use a generic wrapper rather than dedicated, credential-verified adapters.

Does vibelive provide a managed relay?

vibelive runs its relay on the host machine. The package does not provide a managed remote relay service.

How is vibelive different from tmux?

vibelive shares one wrapped process with a driver queue and chat. tmux manages terminal sessions, windows, and panes.

How is vibelive different from tmate?

vibelive uses a local/LAN WebSocket and explicit driver arbitration. tmate provides SSH-based terminal sharing.

Is vibelive end-to-end encrypted?

vibelive ships plain WebSocket transport. An encrypted remote relay is roadmap work.

Can two people type into the agent at once?

vibelive allows one driver to send input. Other participants can chat and queue for control.

Does vibelive work with full-screen terminal apps?

vibelive uses child-process pipes without PTY semantics. Raw-mode, resize, and full-screen terminal behavior are limited.

Do late joiners see earlier output?

vibelive sends retained output history before live updates. The default rolling buffer keeps up to 5,000 chunks.

License

MIT.

Frequently asked questions

What is io.github.pooriaarab/vibelive?

Multiplayer for agentic coding terminals: live cursors, presence, chat.

How do I install io.github.pooriaarab/vibelive?

It runs on your machine. In Claude Code: claude mcp add vibelive -- npx -y vibelive@0.2.2 mcp. For other clients, paste the JSON configuration from this page into their MCP settings.

Does io.github.pooriaarab/vibelive need an API key?

Its registry entry lists no required key or secret. Some features may still need an account with the service it connects to.

Is io.github.pooriaarab/vibelive open source?

Yes. Its code is on GitHub under the MIT license.

Who maintains io.github.pooriaarab/vibelive?

It is published by pooriaarab. AI.tn lists it but does not maintain or endorse it.

From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.