MCO CLI
A CLI that runs Claude Code, Codex, Gemini CLI and other agents in parallel on one task and collects their answers side by side
- 552 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- `mco run` can write to the workspace, and the `yolo` mode removes the chosen agent's approvals.
- `--allow-paths` only validates the requested scope and does not replace an OS sandbox.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/mcoDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- `mco run` can write to the workspace, and the `yolo` mode removes the chosen agent's approvals.
- `--allow-paths` only validates the requested scope and does not replace an OS sandbox.
This is third-party code. Review the repository files before installing.
What it does
MCO installs as the `@tt-a1i/mco` package and drives several coding agents as a working set. `mco review` runs a read-only repository review, and `mco run` executes a task, including with write access to the workspace. Executors are chosen explicitly with `--providers`: Claude Code, Codex CLI, Gemini CLI, OpenCode, Qwen Code, Copilot CLI, Hermes, Pi, Grok Build, Cursor CLI and Antigravity. You can also install the bundled `mco-cli` skill: `npx @tt-a1i/mco@latest install` puts it into the agents you choose. Such an agent reads `mco -h`, shows you the set of executors and runs the task only after you confirm. Every answer is kept whole, as the agent returned it. One executor failing does not erase the answers of the others. Tasks run in parallel, in a chain or split by files, and `--debate` starts a discussion with a final synthesis. Permissions follow a profile: `read_only`, `write` or `yolo`.
Who it is for. Developers who compare code reviews from several models and groups where Claude Code should hand part of the work to Codex or Gemini CLI.
Good fit when
- Before a release you want two or three models to look at the weak spots of a repository
- Agents disagree and you need to compare their full answers, not a short verdict
- A task splits by files and each agent takes its own slice without overlap
- You want a preview with `--dry-run`: who will run and with which permissions
- An agent with the `mco-cli` skill distributes the task between executors itself
Not a fit when
- Only one agent is in play and there is nothing to compare
- You want one scored result with consensus: MCO returns raw answers without conclusions
- You need isolation: `--allow-paths` validates the requested scope and is not an OS sandbox
- You need separate git worktrees for parallel edits: MCO does not create them
Example request
Use MCO to run a security review with Claude and Codex and an architecture review with Pi, Show me the commands and the set of executors first.Limitations
- Each agent must be installed and signed in separately: MCO only calls the CLIs you have.
- The `yolo` mode switches on the broadest approval-bypass profile of the chosen agent. It needs an explicit flag.
- For parallel writes you split `--target-paths` yourself, otherwise edits collide.
- Installation needs Node.js for `npx` and Python 3.10 or newer for MCO itself.
- Hermes in oneshot mode bypasses approvals, so it requires an explicit `--execution-mode yolo`.
How to disable. Delete the `mco-cli` skill folder from the calling agent's skills directory and run `npm uninstall -g @tt-a1i/mco` if you installed it globally.
FAQ
How does `mco review` differ from `mco run`?
`mco review` runs in `read_only` mode and does not change files. `mco run` executes a task and by default gets write access to the workspace.
How do I see which agents MCO finds on my machine?
Run `mco doctor --json`: it reports which CLIs are installed. MCO never picks executors from detected binaries, you name them yourself.
What does `--dry-run` do?
It prints the launch plan without executing: executors, models and the access profile. Run it before a writing `mco run`.
Can I pin a model for one run?
Yes, `--provider-models-json` takes JSON, for example `{"codex":"gpt-5.4"}`. The CLIs' own settings stay unchanged.
How do I call MCO from Claude Code or Codex?
Install the skill with `npx @tt-a1i/mco@latest install --agent claude-code --agent codex --yes`. From then on the agent works from the `mco -h` help.
Discussion
No questions or reviews yet.
Related
CLI toolsWork modes for AI coding agents: a light solo mode for daily tasks and /man with a plan, checks and review for big changes
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A CLI that gives six coding agents project context, over 80 skills, verification and resumable tasks across sessions
Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
CCManager is a session manager for Claude Code, Codex, Gemini CLI and other agents: each runs in its own Git worktree with status and hooks
One desktop window for Claude Code, Codex, Gemini CLI and OpenCode: terminals, pull requests, diffs and approvals
README in short
The quick start, the built-in executors, access profiles and output modes (text, JSON, JSONL and Markdown files) are covered. Separate guides in `docs/` describe flags, configuration and error contracts.