ProtoPie MCP went official. Interaction is the taste test now
AI can draw your app in seconds. It still can't feel how it moves.
I wrote after DevDay that the Codex to Figma roundtrip punishes messy systems twice — every weak token travels in both directions. ProtoPie just made that problem concrete for motion. Its MCP release connects agents to the part of the product most founders fake: the interaction underneath the screen.
ProtoPie MCP connects AI agents to real prototypes over MCP — Studio MCP builds and edits with your library, Code MCP turns finished behavior into code — and solo founders who prototype the interaction ship handoffs that hold.
What did ProtoPie MCP actually ship?
Two servers, two jobs, one workflow from prompt to code.
ProtoPie's announcement making MCP official frames it plainly: build with AI, refine in ProtoPie, move to code with context intact. Studio MCP runs inside Studio and edits what's open — layers, triggers, responses, variables — from a chat prompt. It pulls from your own libraries, so new screens arrive as instances of your components instead of fresh one-offs.
Code MCP does the second half. ProtoPie's guide to Code MCP turning prototypes into framework code shows it as read-only and link-based: paste a Cloud link into Claude Code or Cursor and the agent reads scenes, components, formulas, and interaction logic, then writes React, React Native, Flutter, SwiftUI, or Compose code from it. Dev View sits between them — pick a scene or a single layer and hand the agent exactly that slice.
Transparency first: I haven't run Code MCP on a client build yet — I'm reading the docs and release notes. What stands out is the wire. Both servers run on the Model Context Protocol, the open standard for agent tool access, the same plumbing Figma and Codex use. The industry picked a wire. ProtoPie just plugged motion into it.
Why does interaction decide the handoff?
Because pictures transfer fine. Behavior doesn't.
Think about your last handoff that went wrong. The screens matched — then QA found the modal with no dismiss path and the loading state that never clears. None of that lives in a screenshot. It lives in triggers, conditions, and variables.
Here's my opinion, stated flatly: a static mock is a promise your code has to guess at. An interactive prototype is the answer key.
The library detail matters more than the AI detail. An agent inventing buttons from a prompt gives you three variants by lunch. An agent instancing your library gives you a screen to hand off.
How do you run one flow through the loop?
You pick one flow users pay for, and you keep the agent on a short leash.
- One flow, not the app. Pick checkout, onboarding, or the core create step. The flow with state — loading, empty, error, success — is where this pays off.
- Library only. Turn on your ProtoPie libraries before you prompt. If the agent can't instance it, it can't ship it.
- Target by ID. Copy the element ID for the layer or trigger you want changed and paste it into the prompt. "Change the interaction for [element ID]" beats three paragraphs of description.
- Hand off a slice. Don't send the whole file. Copy a Dev View link for one scene, ask for one framework, and compare timing and state line by line.
Run that loop once. If the generated code keeps your easing, conditions, and variables without a rewrite, your system held. If it smooths everything into defaults, fix the prototype before you blame the model.
When should you stay in the static mock?
When the question is still what, not how it moves.
Early exploration doesn't need triggers. Copy direction, hierarchy, and "should this screen exist at all" are judgment calls — they need you with a delete key, not an agent wiring taps between screens you might cut tomorrow.
My rule: if you're still deciding what the product says, stay in sketches. Once you've decided and need to prove how it feels, move into ProtoPie and let the agent do the wiring. Motion too early just makes a bad idea feel finished.
The agent handles the wiring. You decide what's worth feeling.
Frequently asked questions
Two MCP servers that connect AI agents to ProtoPie. Studio MCP lets agents like Claude Code, Cursor, and Codex read and edit the prototype open in Studio by prompt. Code MCP gives coding agents a read-only link to a finished prototype — scenes, layers, variables, and trigger-response interactions — so they can rebuild it in React, React Native, Flutter, SwiftUI, Compose, Qt/QML, or ArkTS.
About the author
mosh
mosh is a product designer for growth, working with design thinking and ever-improving design systems. What matters: fixing conversion, whether in B2B dashboards or direct-consumer apps.
Keep reading
- Escape Ludo: game design
A covid quarantine project. Sprite design, character formation, levels, and inventory UI for a multiplayer Ludo game.
- Go Laundry: service booking platform
One platform to book cleaning services in Qatar. Art-directed brand mascots, app UI, and UX research.
- SaaS activity tracking: designing an admin log teams trust
Admins needed to see every team action in one place. Research with two managers and eight user stories turned a confusing log into filters, search, share, and export.