Agent Orchestra — Development skill for Claude Code

agent-orchestra

🚧 **Work in progress** — personal sandbox for exploring multi-agent orchestration patterns.

This repo is my personal playground where I'm experimenting with recent agent frameworks and orchestration patterns that have appeared in late 2025 / early 2026:

• Claude Agent SDK (Anthropic)
• LangGraph — stateful multi-agent workflows
• Model Context Protocol (MCP) — standard tool interface across agents
• Computer Use / browser agents — follow-up experiments planned

The goal is not to ship a product. The goal is to *understand the trade-offs* between different orchestration patterns (sequential chain, parallel fan-out/fan-in, supervisor, hierarchical) and to get comfortable with the tooling.

Status

Area	Status	Last touched
Basic agents (Python)	✅ Done	2026-03-28
LangGraph supervisor workflow	✅ Done	2026-04-02
Sequential chain experiment	✅ Done	2026-04-05
Parallel fan-out experiment	✅ Done	2026-04-08
MCP server (TypeScript)	🟡 In progress	2026-04-11
Hierarchical orchestration	🟡 In progress	2026-04-16
Computer Use integration	⬜ Planned	—
Benchmarks / evals	⬜ Planned	—

See [`docs/ROADMAP.md`](docs/ROADMAP.md) for details.

Layout

agent-orchestra/
├── python/            # Agents + LangGraph workflows
│   ├── agents/        # Individual role agents
│   ├── graphs/        # LangGraph workflow definitions
│   ├── tools/         # Tools exposed to agents
│   └── main.py
├── mcp-server/        # TypeScript MCP server exposing agents as MCP tools
├── experiments/       # Standalone experiments, one per pattern
└── docs/              # Notes, architecture, roadmap

Quick start

# Python side
uv sync
cp .env.example .env   # add your ANTHROP