Dyad LogoDyad

AI-reviewed MCP tool consent, per-app localhost preview domains, and an experimental shell for Pro Agent mode

Dyad v1.18.0 Beta 1 helps Pro users reduce MCP consent prompts, adds opt-in localhost domains for individual apps, and introduces an experimental safety-reviewed shell for Agent mode.

Highlights

Pro users can now use an AI reviewer to automatically approve MCP tool calls that are determined to be safe.

This reduces consent prompts while preserving a review step before tools run. Calls that are not approved automatically continue through the normal consent flow.

Per-app localhost preview domains (Experimental)

You can now give each app its own localhost preview domain instead of using a shared localhost address.

Dedicated domains help isolate browser data such as cookies and local storage between apps. This is particularly useful when developing multiple apps with authentication.

Enable the preview-domain experiment in Settings to use this feature.

Per-app localhost preview domains are experimental. Some integrations may need their allowed URLs or callback URLs updated.

Safety-reviewed shell for Pro Agent mode (Experimental)

Pro users can opt in to a new shell experiment for Agent mode. This lets Agent run shell commands when needed, with commands checked by a safety reviewer before they can run automatically.

Shell access is experimental. Review your app changes and command output carefully, and please report any issues.

Enhancements & fixes

  • Added support for PDF attachments as chat context.
  • App test runs are now queued instead of cancelling an active run.
  • Questionnaires now remain open until you respond or dismiss them.
  • Catalog plugins can now include optional setup inputs.
  • Made model usage selection easier to see.
  • Added a Refresh button for detecting updated Wrangler configuration in the Cloudflare publishing setup.
  • Fixed the Cloudflare repository access prompt so its controls remain usable.
  • Chats and app tests now continue running during Supabase function deployments.
  • Preserved new Anthropic thinking content when switching chat modes.
  • Streaming responses no longer scroll you back to the bottom after you have scrolled away.
  • Fixed Claude pull request reviews losing write access after changing directories.
  • Applying a template in place now preserves existing .env files, .git, and .dyad data.
  • Improved command launching reliability on Windows.
  • Improved Coolify setup validation and handling when switching between instances.

Feedback

If you run into issues or have suggestions, please help us improve by reporting a bug.

On this page