Project: Creative Apps (GitHub Copilot) - MCO (Multi-CLI Orchestrator)

Track

Creative Apps (GitHub Copilot)

Project Name

MCO (Multi-CLI Orchestrator)

GitHub Username

tt-a1i

Repository URL

https://github.com/mco-org/mco

Project Description

MCO is an open-source multi-agent orchestration layer for AI coding workflows. Instead of relying on a single coding assistant, MCO dispatches one task to multiple agent CLIs in parallel (Claude Code, Codex CLI, Gemini CLI, OpenCode, and Qwen Code), then aggregates outputs into actionable results.

It solves a practical engineering gap: single-agent workflows are fast but narrow. Different models catch different issues and propose different solutions. MCO enables a team-style workflow: assign once, execute in parallel, deduplicate findings, and optionally synthesize consensus.

MCO supports two modes: mco run for general tasks and mco review for structured code review. It also provides mco doctor for environment checks, --format markdown-pr for PR-ready comments, --format sarif for code scanning workflows, --synthesize for consensus summaries, and --include-token-usage for optional usage visibility.

Demo Video or Screenshots

Demo Video: https://youtu.be/18wsOMoHunI (Chinese narration with bilingual subtitles)
Screenshots: https://github.com/mco-org/mco/tree/main/docs/assets
Live Demo: N/A (CLI tool)

Primary Programming Language

Python

Key Technologies Used

• Python 3.10+ (core runtime and orchestration)
• Node.js wrapper package (@tt-a1i/mco) for npm distribution
• Provider adapter architecture for multi-CLI integration
• JSON / SARIF / Markdown outputs for automation and CI
• GitHub Copilot for coding acceleration, refactoring, and docs iteration

Submission Type

Individual

Team Members

N/A

Quick Setup Summary

1. git clone https://github.com/mco-org/mco.git
2. cd mco
3. Install: npm i -g @tt-a1i/mco (or python3 -m pip install -e .)
4. Verify: mco doctor --json
5. Run:
   • mco run --repo . --prompt "Summarize this project in 3 sentences." --providers claude,codex,gemini --json
   • mco review --repo . --prompt "Review for high-risk bugs." --providers claude,codex,qwen --format markdown-pr

Technical Highlights

• Unified adapter contract across five coding-agent CLIs with consistent detect/run/poll/cancel behavior
• Parallel wait-all execution model where one provider failure does not block others
• Cross-provider finding deduplication with detected_by provenance
• Optional synthesis pass (--synthesize) for consensus/divergence summary
• Multi-format outputs (report, json, markdown-pr, sarif)
• Optional token usage extraction (--include-token-usage)

Challenges & Learnings

• Provider CLIs produce different output styles (plain text vs JSON event streams), so robust extraction/normalization was required.
• Binary resolution and auth probing can diverge from runtime PATH semantics; detection and runtime paths must stay aligned.
• Multi-agent output can be noisy; synthesis + dedup significantly improves actionability.

Contact Information

Optional

Country/Region

China

Copilot Usage (Creative Apps relevance)

GitHub Copilot was used during implementation and iteration for adapter refactors, CLI UX updates, test drafting, and documentation polishing. We used Copilot suggestions as accelerators, then validated behavior through local runs and the Python test suite before merging.

This issue is a project submission describing MCO, an external repository at https://github.com/mco-org/mco. It names no file, test, or change to make in the agentsleague repository, so a newcomer has no concrete implementation entry point or completion criterion.