Use Claude Code, Codex, and Cursor in one structured session.

The manual relay problem

A common workflow starts with a plan in Claude Code, a review in Codex, and another pass in Cursor. Each tool can add value, but the developer becomes the transport layer: copying context, translating objections, and deciding which message represents the latest plan.

This manual process is difficult to reproduce and easy to lose. An agent may respond to an outdated version, a disagreement may disappear from the final summary, or a provider failure may be mistaken for agreement.

How Codebate coordinates the tools

Codebate launches supported provider CLIs from a local workspace and gives them roles within the same task. Agents can propose, review, challenge, and converge through a structured control contract rather than an untracked chain of copied messages.

Claude Code and Codex CLI are supported providers. Cursor CLI integration is experimental and intentionally conservative. Exact capabilities can differ by provider, operating system, and configured safety mode.

• Shared task and repository context
• Provider-specific adapters and safety rules
• Bounded discussion rounds
• Explicit open items and user decisions
• Structured transcript and official outcome

Subscriptions and authentication

Codebate is open source and does not include model access. You continue to use the provider accounts, subscriptions, and CLI authentication you already have. This avoids introducing another AI model subscription, but provider usage limits still apply.

Choose roles, not just brands

The strongest workflow is usually not three agents editing at once. Assign one agent to write, another to review correctness, and another to challenge architecture or test coverage. Different roles create useful independence and reduce conflicting writes.