OpenCode now runs on your Claude Code login

OpenCode now runs on your Claude Code login and its models. The new built-in provider is called Claude Agent. On desktop, setup checks that Claude Code is installed and signed in.

Slide text

anomalyco/opencode · release notes · PR #58782 · PR #60603

OpenCode now runs on your Claude Code login

New built-in provider: Claude Agent

Desktop setup checks Claude Code is installed and signed in

Run /goal with an objective, and the session keeps working. It stops when a reviewer decides the objective is met. You can pause, resume or clear it in the terminal app or on desktop.

Agents can now create and edit images. It turns on by itself when a connected provider has an image model, including GPT Image 2 on a Codex login, MiniMax and Meta Muse.

Two new experiments stay off until you turn them on. Scheduled prompts wake a session on a timer or a cron, and now fire under opencode serve. Artifacts let agents publish and serve HTML pages.

On desktop, terminals belong to the session and survive a server restart. Long sessions do much less rendering as replies stream, and new keyboard shortcuts pick a worktree, its base branch, or shell mode.

Behind a proxy, the terminal app no longer hangs restarting its own service. MCP servers that send messages over ten megabytes stay connected, and a full context window on LiteLLM now compacts instead of stopping.

Three changes can break things. Terminal plugins need OpenTUI 0.8. Under managed policy, MCP servers, skills and plugins get their own rules. And code that imports the AI package loses four providers, including Replicate.

If you pay for Claude Code, try Claude Agent first. It brings your login and its models straight into OpenCode. The upgrade steps are in OpenCode's release notes on GitHub.