DIVE vs Drawbridge

Drawbridge captures annotations and drops task files into your project folder; DIVE edits the live page and hands off over MCP, no folder binding, so it works across parallel agents and worktrees.

Both turn visual feedback into work for an AI coding agent. Drawbridge is annotate-and-queue: you comment or draw a box, it writes screenshots + `.moat/` task files into one connected project directory, and the agent reads them via a `/bridge` command. DIVE is a live two-way loop: you edit the page directly (or pin element/region comments), and the agent pulls the diff over MCP and pushes changes back, then marks your comments resolved.

Pick DIVE when

You run multiple agents or git worktrees at once, DIVE's MCP attaches per session with no folder binding, so nothing pins you to a single directory or risks writing tasks into the wrong tree.
You want to actually edit the design (typography, colour, layout, motion, effects), not just describe the change.
You want the loop to close: the agent applies changes to the live page and marks your comments resolved (mark_comment_resolved), instead of a one-way file drop.
You want a persistent, searchable Changes history and CSS / Tailwind / SCSS / JSX export.
You'd rather not grant a browser persistent write access to your project folder.

Pick Drawbridge when

You never run the MCP companion and want a zero-server flow, Drawbridge writes plain files an agent reads on its own.
You want annotations checked into git as `.moat/` task files for an async, review-later workflow.
Drawbridge's `/bridge` step/batch/yolo command already fits your team's habits.

Feature	DIVE	Drawbridge
Visual editing on a scriptable webpage	Yes, typography, colour, layout, spacing, motion and effects on localhost, staging or production pages the extension can script. Browser-protected pages are excluded.	Annotate only, comment pins + freeform rectangles, no live editing
MCP (Model Context Protocol) handoff to AI agents	Yes, Cloud, Local, and Self-hosted modes; eight common session tools, plus Local feedback rounds	No, writes .moat/ task files via the File System Access API
Persistent change history (Changes tab)	Yes, searchable, filterable, exportable	Task files (to do / doing / done) in the connected folder
Open source	Yes (MIT)	Check current licence
Price	Free MIT-licensed extension	Check current pricing
Markdown / JSON export of the diff	Yes	Markdown + JSON task files
Best fit for	Designers, developers, QA, PMs, content, indie hackers, agencies, vibe coders	Annotate-and-queue for AI coding agents

The honest take

Drawbridge is a clean, focused tool, and its no-server file handoff is genuinely simpler if you never run a companion: annotations land as git-friendly task files an agent reads on its own. The trade-off is the folder binding, it persists one project directory handle, which gets awkward when you're juggling parallel agents or worktrees, and it captures intent rather than letting you make the change. DIVE deliberately skipped file handoff for exactly that reason and leans on a live MCP connection instead. If you want async, checked-in task files, Drawbridge fits; if you want to edit directly and close the loop live across many sessions, DIVE does.