The review surface for agent-built prototypes.

Your agent builds a prototype, your team comments on any element, and your agent rebuilds it. Like Claude Design or Figma Make — but for any agent.

get started →
Bring your own agent
Claude Code
Codex
Cursor
Windsurf
Cline
opencode
Copilot
grocer-checkout v4 · edited 2m ago BSSKAI
Pages4
1Browse
2Cart2
3Pay✓
4Confirm
＋ add page
Page brief✎ edit
BS Björn · PM

Last step before payment — keep it scannable so the total never feels like a surprise.

Main risk is drop-off when cost feels unclear, so checkout should read as the obvious next step. Sara wants an editable quantity stepper per item — agree, let's try it.

68%
9:41●●● ▮
Your cart
2 items, ready to go
Olive harvest box500ml · cold-pressed
$24
– 1 +1
Sea salt flakes120g · Maldon2
$11
– 1 +
Subtotal$35.00
DeliveryFree
Checkout · $353
Comments2 open · 1 resolved
BSBjörn2m
Make checkout the clear primary action — full width, larger. It reads like a secondary link right now.
3 anchor #c7d2 · open
AIAgentnow
Read the Cart brief + 2 open comments — enlarging checkout, adding a quantity stepper, preserving anchors.
update_element() · v4→v5
SKSara8m
Can we show a quantity stepper on each line item?
1 anchor #a3f9 · open
SKSarav3
Show the unit size under each item name.
2 anchor #b1c0 · resolved in v3
Reply or pin a new comment…
↑
How it works
1
your agent
Builds the prototype. Renders each screen over MCP and tags every element with a stable anchor id.
2
your team
Comments on anything. Pin feedback to a specific element — not a vague "make the button better."
3
your agent
Rebuilds — anchors survive. Reads the unresolved comments, rewrites the page, and every comment stays put.
How it's different

Bring your own agent — no lock-in. Claude Code, Codex, Cursor, or anything that speaks MCP drives the loop, on the model and token budget you already pay for. No forced model, no metered credits, no per-seat pricing.

opencanvas
Claude Design · Figma Make
Agent
Bring your own — any MCP client
Their built-in agent only
Model
Any model you choose
Locked to their model
Cost
Your agent's own budget — no token markup
Metered credits & token budget
Your data
Versioned & exportable, in your repo
Locked in their cloud
FAQ
What is opencanvas?+
opencanvas is a review surface for agent-built prototypes. Your agent builds a prototype over MCP, your team comments directly on any element, and your agent rebuilds it — with every comment anchor surviving the regeneration.
Which agents and models does it work with?+
Any MCP client — Claude Code, Codex, Cursor, Cline, Windsurf, opencode, GitHub Copilot, and more. opencanvas is model-agnostic, so you bring your own agent and run on whatever model you choose. There's no built-in agent and no forced model.
How is it different from Claude Design or Figma Make?+
Those tools lock you to their built-in agent, their model, and their metered token or credit pricing. opencanvas is MCP-native and model-agnostic: you bring your own agent, run on the token budget you already pay for, and keep your data versioned and exportable in your own repo.
What happens to my comments when the agent rebuilds a prototype?+
Comments are pinned to specific elements through stable anchor IDs that survive full DOM regeneration. When the agent rewrites a page, feedback threads stay attached to the right elements and travel across every version.
Is there any token or usage lock-in?+
No. opencanvas runs on your own agent's budget — there's no token markup, no metered credits, and no per-seat pricing.
Do I own my data?+
Yes. Snapshots are versioned and exportable, and you can track them in your own repo. Nothing is locked inside a vendor cloud.
Do I need to log in? How is access secured?+
Not yet — during early access opencanvas has no login or authentication. Each project lives at a non-guessable URL, and editing requires a secret key passed in the URL's query string. Anyone with the view link can see a project, and anyone with the edit key can change it — so treat your links like passwords and don't put confidential data in yet. Full authentication is coming soon.
Is opencanvas available yet?+
Yes — opencanvas is live as a v0.1 beta. Get started by connecting your agent over MCP.