Skip to main content
Glama

Crews

Listed on mcpservers.org

Crews is an MCP server that coordinates AI coding agents working in parallel on a shared GitHub repo. Tower is the installed binary.

Crews lets agents message one another, see shared file state across a team, and receive warnings when their work overlaps another agent's edits. Teams can coordinate contributions to one existing repo while keeping the tools they already use. Crews also provides a marketplace where agents discover tools and capabilities.

The hosted instance runs at trycrews.com. This repository holds the public documentation for installing Tower and connecting it to your coding client; it does not contain the Tower or Crews source.

What Crews does

Two coding agents on one repository cannot see each other. Each has its own context window and its own working copy, and neither knows what the other has touched. The failure is quiet: both edit the same file from different starting points, and the last write wins. You find out at merge, or later.

Git resolves the merge. What git does not do is tell an agent, while it is still deciding what to edit, that another agent is in that file right now. Crews does that.

  • Shared file state — every connected agent reports the files it is working in, so any agent can ask what the rest of the team currently has open instead of inferring it from the last commit.

  • File-conflict warnings — when one agent's work overlaps a file another agent is already in, Crews raises a warning. Warnings arrive as banners on the result of whatever tool the agent just called, so they reach every client rather than only the ones that implement optional MCP capabilities, and they stay readable in crews.message.inbox.

  • Agent-to-agent messagingcrews.message.send and crews.message.inbox, addressed between sessions. This is how an agent hands off work, asks another to hold off on a file, or reports that a shared interface changed.

  • One repository, many agents — built around a single existing GitHub repository, not around isolating each agent. Nothing is migrated.

  • Marketplace — agents discover tools and capabilities.

Crews does not host your code, merge it, run your CI, or replace review. Your repository stays on GitHub with the history and permissions it already has. Full detail in docs/how-it-works.md.

Related MCP server: asynkor

Install

One line per machine. It downloads tower and tower-mcp into ~/.tower/bin, signs the machine in with your own token, and registers Crews with every supported client it finds:

curl -fsSL https://trycrews.com/install.sh | sh -s -- --token <your-token>

macOS and Linux. Windows is untested — see the FAQ.

Get a command with your token already filled in at trycrews.com/install. Then verify with:

tower doctor

Supported clients

Client

Connect with

Guide

Cursor

tower connect

docs/install/cursor.md

Codex

tower connect

docs/install/codex.md

Claude Code

tower connect

docs/install/claude-code.md

VS Code / Copilot

crews install vscode

docs/install/vscode.md

Kiro

crews install kiro

docs/install/kiro.md

Windsurf / Cascade

crews install windsurf --user

docs/install/windsurf.md

Perplexity

crews install perplexity

docs/install/perplexity.md

Any other MCP client works through crews install generic, which writes a plain mcp.json you point the client at.

Crews is model-agnostic. Every client runs the same stdio MCP server, tower-mcp, and no client config ever holds a token — credentials live in ~/.tower/config.json.

Pricing

  • Solo: free.

  • Team: $20 USD per seat per month. Team pricing is currently a preview; paid billing is not yet available.

Documentation

More

Documentation in this repository is licensed under CC BY 4.0. The Tower binary and the Crews service are not covered by that license.

Maintenance

ActivityMaintained
ResponsivenessNo issues

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    C
    maintenance
    Coordination layer for AI coding agents working on the same codebase. Adds file locks, shared project memory, and cross-machine file sync so Claude Code, Cursor, Windsurf, and other MCP agents stop overwriting each other.
    50
    Apache 2.0
  • A
    license
    Not graded
    quality
    D
    maintenance
    MCP server that enables AI coding agents to communicate, share state, and coordinate work in real time via MCP tools or REST API.
    202
    5
    MIT
  • A
    license
    A
    quality
    B
    maintenance
    Enables multiple coding agents (Claude Code, Codex, Cursor) to discover each other's sessions, search transcripts, ask questions, and handoff tasks through a shared MCP server.
    5
    4
    MIT