Terminal for Agentic Development — Developer Documentation
Released: 2026-04-03
You can now type in the terminal while Claude is thinking. When Claude Code is waiting for input or processing, Ocean passes your keystrokes through to the shell instead of buffering them. This means you can run quick commands, check git status, or browse files without waiting for the agent to finish.
Ocean now exposes an MCP server that AI models can call directly. The Agent API lets external tools query session state, trigger merges, and read conflict data — turning Ocean into a programmable backend for agent orchestration workflows.
A new fetch_remote utility runs automatically before push operations, ensuring your local branch is in sync with the remote. This prevents failed pushes due to diverged history and reduces manual git pull steps in agent-driven workflows.
New context menu options for workspace management:
A comprehensive security review addressed critical and high severity issues:
test-all.sh script with detailed per-suite reporting