Claude Agent Teams is a free, open-source desktop application that turns AI agents into a coordinated team. You define the roles, and agents handle tasks autonomously — writing code, reviewing each other's work, and messaging teammates — while you watch progress on a real-time Kanban board.
The app supports Claude, Codex, and OpenCode runtimes with access to 75+ LLM providers. You don't need API keys if you already have a subscription — the app auto-detects your existing provider access. Each agent gets its own task stream, logs, and branch strategy (single branch or git worktree per agent).
What sets Claude Agent Teams apart is the built-in review workflow. Agents review each other's code changes, and you can accept, reject, or comment on individual hunks — similar to how code review works in Cursor. The Kanban board shows five columns (Backlog, To Do, In Progress, Review, Done), with tasks moving automatically as agents work through tasks.
The v1.2.0 release added a real-time Agent Graph — a force-directed network visualization showing how agents communicate, with animated message particles and cross-team ghost nodes. This gives you a visual map of which agents are collaborating and how information flows through your team.
Other notable features include MCP integration for connecting external tools and agent plugins, deep session analysis with six-category token tracking (user messages, Claude.md instructions, tool outputs, thinking text, team coordination), and auto-resume after rate limits with cooldown nudges so work continues even when providers throttle.
For developers who prefer working with a single agent, the solo mode creates a self-directed agent that generates its own tasks and shows live progress on the board. This saves tokens compared to a full multi-agent team while still providing the Kanban visibility and review workflow.
The app includes a built-in code editor with Git support, so you can make quick edits without switching windows. Branch strategies let you choose between a single shared branch or git worktrees per agent to keep changes isolated. The notification system is configurable — get alerts for task completion, agent input requests, comments, and errors.
Task creation supports attachments. Send a message to the team lead with attached images and the lead auto-creates tasks with files. The post-compact context recovery feature restores key team-management instructions after the runtime compacts the context window, keeping agents aligned even during long sessions.
Built with Electron 40, React 19, TypeScript 5, and Tailwind CSS 3. Licensed under AGPL-3.0. Available on macOS (Apple Silicon and Intel), Windows, and Linux with pre-built binaries. No credit card or API key required to get started — just download and launch the setup wizard.