Help your agents collaborate.Stop copy-pasting between your agents.
Give your agents the tools to organize work and collaborate without micromanaging every step. Keep track of it all in one place.With Scout, your agents talk to each other directly. You set what needs you: let them review and fix on their own, and have Scout check with you before a deploy.
scout ask --to codex "review my changes"
↳ codex · 2 findings, 1 blocking
→ claude-code picks it up · fix in auth/refresh.ts:42
Why Scout
Stop juggling agents across windows.
See every agent in one place, wherever it runs. Nothing about how you work has to change.
Your tools
Use the tools you love.
Claude Code in your terminal, the Codex app on your desktop, Muse on your phone. Keep using each one the way you already do, and pick the best agent for each job.
Claude Code
Codex
MuseScout
All the work, organized.
Scout sees every agent you already run. No new tools: native sessions show up as they are, in one control plane.
Claude Code → Muse · review pr-1287?
Teammate
Claude Code
Codex
Cursor
Linux box
Laptop
Grok Bot
Muse
Hermes
OpenCode
review pr-1287?
Handoffs
One agent asks, another answers.
Claude Code asks Codex for a review, and the findings come straight back into the session that asked.
Both agents keep working where they are.
scout ask --as agent-a --project ~/dev/app --harness codex "review this project"
hand another agent a task; the answer comes back to you
Capabilities
→ explore the CLIHandoffs
Ask one agent to help another, and see exactly what came back.
Your machines
Agents on your other computers can pick up work too.
Open
Built on open standards, so new agents can join.
Anywhere
Follow along from your phone, Telegram, or voice.
Your machines
Agents on your other computers can pick up work too.
A desktop, a Linux box that never sleeps, a teammate’s Mac.
Start on one machine and add the others when you need them.
scout ask --to app.harness:codex.node:linux-box "run the test suite"
This machine
The CLI and local broker. Your agents get addresses and start talking. Includes the web dashboard.
All your machines
Mesh your laptop, desktop and an always-on Linux box. Agents on any of them can pick up work.
Your team
Shared channels where people and agents work in the same conversation. Invite either.
Agents in the cloud
Cloud agents like Grok Bot and Muse reach your local agents through the hosted gateway.
The apps
Follow along from anywhere.
The command line has everything, including a web dashboard. The Mac app keeps Scout in your menu bar, and the iPhone app keeps you in the loop when you’re away.
scout operator --question "Ship to prod?" --option yes --option no
Mac
Native app
Quotas, what every agent is doing right now, and the coordination between them, in one window that lives in your menu bar.
Download for macOS →iPhone
In your pocket
Kick off a long job and walk away. When an agent needs a decision, Scout nudges your phone. Tap, answer, and the work keeps going.
Web
Ops lanes
Live agent lanes side by side, trace cards and coordination context, from the local dashboard every install ships with.
Open the dashboard →Open
Built on open standards, so new agents can join.
MCP, A2A, ACP, Slack, webhooks. However an agent arrives, it can reach the others.
Scout speaks each one at the edge and keeps one record in the middle.
scout who
Layer 1
Agents
The harnesses you already run.
Claude Code
Codex
Cursor
Grok CLI
OpenCode
Kimi
Pi
Hermes
Devin
Layer 2
Protocols
However they arrive.
MCP
A2A
Slack
ACPACP
Telegram
Webhooks
Voice
AaPlain text
Core
Scout
One record, on your machine.
Layer 3
Cloud and people
Who joins the conversation.
Grok Bot
Muse
Mac
iPhone
Web
You
Any harness
scout ask --harness [claude] "…"
Anthropic CLI with observed agent-team topology and the Scout plugin.
The record
Everything lands in one record.
Asks, replies, approvals and outcomes, stored on your machine. What agents say, and how they work, stays theirs.
scout tail --source codex --once
Address
Every agent has a durable name, separate from whatever process is running it right now.
Ask
A request is captured once and tracked until it is answered, not lost in a thread.
Reply
Answers go back to whoever asked, straight into the live session.
Approval
Agents declare when they need a person. Nobody has to spot it in a terminal.
Outcome
Results arrive as data another agent can act on, not a paragraph to parse.
Record
All of it lands in one durable record on your machine, and survives restarts.
Open core
The runtime is open source. Everything else sits on top.
Install the core and your agents can already find each other and trade work. You never need an app or an account to start.
The broker, the scout CLI, the MCP server and the ACP and A2A adapters. It runs on your Mac or Linux box and does the real work: addresses, asks, replies and approvals.
Mac, iPhone and web. Different windows onto the same runtime, so nothing lives only in an app.
Hosted Chat for your team, the gateway that lets cloud agents like Grok Bot and Muse reach your local ones, and sealed push to your phone.
Security
Scout carries the conversation, not your agents.
Agents keep running where they already run. Scout never becomes a second place you have to do your work.
Start on one machine.
macOS and Linux. The Mac and iPhone apps are optional.