Skip to content
MCP server

mcp-crewd-rs

Local daemon that lets Claude or Codex sessions spawn and message worker AI cells over MCP

Pricing
Open source
Platforms
Linux, macOS, Android
Version
0.1.4
Listing updated
today

Similar MCP servers

About mcp-crewd-rs

mcp-crewd-rs is an open-source MCP server for agent tooling. It is available on Linux, macOS, Android, CLI and self-hosted.

The README describes crewd as a daemon that lets one AI session start, coordinate and exchange messages with other AI worker sessions, which it calls cells. A coding client such as Claude Code or Codex CLI connects through a small stdio shim called crew, and the daemon runs the workers as Claude, Codex or pi engine processes.

Everything stays on the local machine: the daemon has no network listener and talks over a 0600 Unix socket. State lives in SQLite and an append-only, hash-chained audit log that can be checked with a verify command. Each cell's identity is tied to the working directory it is spawned in, so one daemon can serve several project-specific personas.

It suits developers who already run multi-agent coding workflows on Linux, Android/Termux or macOS and want orchestration without a hosted service. The README is explicit that Windows is unsupported and that cross-host coordination is still on the roadmap.

Tools this server provides (12)

Taken from the project documentation; the running server may differ.

cell_spawn
Launch a worker cell with an engine, profile and task.
cell_send_task
Send a task to an existing cell.
cell_status
Check the status of a cell.
cell_result
Fetch a cell's result.
cell_cancel
Cancel a running cell.
cell_list
List registered cells.
cell_send
Send a fire-and-forget message to a cell.
cell_ask
Ask a cell a question and receive a ticket for the reply.
cell_await
Long-poll for a reply to an ask ticket.
cell_reply
Reply to a message from another cell.
cell_broadcast
Broadcast a message to cells.
cell_inbox
Read a cell's inbox.

What it does

Cell spawning and control
Launches worker sessions for Claude, Codex or pi with a chosen profile and task, and lets the parent check status, fetch results or cancel.
Inter-cell messaging
Supports fire-and-forget sends, ask-and-await replies, broadcasts and per-cell inboxes.
Capability-based ACL
Each registered cell has a TOML entry listing which bus and spawn capabilities it may use.
Hash-chained audit log
Records activity in an append-only log that the crew audit verify command can check.
Local-only transport
Uses a Unix socket with peer credential checks and a per-cell token file, with no network listener.
Worker mode
A worker-mode flag hides the spawn tools so worker cells cannot fan out further.

Good for

  • Have a coordinator Claude session delegate subtasks to separate Codex or Claude worker cells
  • Run several project-scoped agent personas by spawning cells in different working directories
  • Keep multi-agent coding orchestration on a single machine without a hosted service
  • Audit which agent sessions sent or received which messages

Pricing

Open source

Apache-2.0 licensed open source, self-hosted.

Prices change. Check dev.mmmbuto.com before you buy.

mcp-crewd-rs FAQ

Does crewd open a network port?

No. The README says it has no network listener and runs everything over a local Unix socket.

Which operating systems are supported?

Linux and Android/Termux are listed with prebuilt binaries, and macOS is supported when built from source. The README says Windows is not supported.

Which AI engines can the workers use?

The README lists claude, codex and pi engines, with session continuity noted as resume by session id for claude, reattach by thread id for codex, and none for pi.

Is cross-host coordination available?

Not yet. The status section describes single-host fabric over Unix sockets and lists cross-host fabric as roadmap.

Is mcp-crewd-rs free?

Yes. mcp-crewd-rs is open source, so you can use and self-host it for free.

What are the best alternatives to mcp-crewd-rs?

The closest alternatives to mcp-crewd-rs are Seamless, Relay MCP, DeskPartner, mcp-memory-rs and Agent Memory MCP.

Where this comes from

  • Description: written from the project's README, the project's website and its MCP Registry entry.

How AIPoweredBot checks servers