Niches · Design workspaces & canvases
Build UI code that matches a design frame
No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.
What counts as done
Code for the named frame in the codebase's framework that uses its existing components and the design's tokens, with image and SVG assets written to disk, and a rendered screenshot a reviewer can compare with the frame export.
Measured, for our own tests: the frame's design data arrives in under 25,000 tokens and three calls or fewer. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: implement this Figma frame
, build this screen from the design
, make the page match the mockup
, turn this frame into SwiftUI
Does the job, with a limit 10
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Figma API + MCP B Figma, Inc. | get_design_context takes a node link, defaults to React and Tailwind but follows a framework prompt; Code Connect maps components; download_assets gives temporary URLs, not files. View and Collab seats get 20 calls a month, and only catalogue clients connect (acc-04, acc-05); acc-03 found it token-heavy, acc-06 succeeded. source |
| Framelink MCP for Figma B Framelink (Graham Lipsman) | get_figma_data takes nodeId and depth and returns compact raw layout and styles for any framework; download_figma_images writes PNG and SVG into the image directory. No component mapping to the codebase and no screenshot tool. The success-without-file bug (acc-25) was closed in April. source |
| Kite D Kite | layer_source returns any layer as HTML with inline styles and layer_export gives SVG, React, Tailwind or HTML with the system's tokens as CSS, so other frameworks are mapped by the agent. No mapping to the codebase's own components, and only for files in Kite. source |
| Magic Patterns B North Park Labs, Inc. | A design is React and Tailwind source the agent reads with legacy_read_artifact_files and ports (acc-40 used it as the spec). No node-scoped design data, asset export, screenshot tool or mapping to the codebase's own components, and MCP needs a paid plan. source |
| pen.dev E High Agency, Inc. | Designs are .pen JSON the agent reads through execute or from git, and the docs' design-to-code flow names the app's own components and tokens. No component-mapping feature, image export only through the CLI, and an open issue says a generic wrong-file error sends agents into retry loops (acc-21). source |
| Penpot API + MCP F Kaleidos Subsidiary SL | execute_code runs plugin API code to read shapes, styles and tokens on the focused page, and export_shape exports assets (limited on remote). Component mapping is by naming only, and the plugin must stay in an active browser tab; acc-19 saw calls fail with an unhelpful error when it wasn't. source |
| Sketch D Sketch B.V. | get_layer_tree_summary takes a layerID and depth, the vendor skill targets React, SwiftUI, Kotlin or others and reuses project components and tokens, get_design_assets writes exports to absolute paths, get_screenshot compares. Needs the Mac app running with the document open, and no mapping feature like Code Connect. source |
| Subframe E Atomic Design Inc | get_page_info returns a page's React and Tailwind code, the CLI syncs Subframe components into the repo, export_image and screenshot_page give images. Output is React and Tailwind only, so a SwiftUI or Vue codebase is ported by hand, and export_image returns a URL, not a file on disk. source |
| UX Pilot MCP server F UX Pilot | read returns a screen and its HTML and design can turn screens into a React app with readable files, so code arrives as HTML or React only. No node-scoped design data, codebase component mapping or asset files on disk are documented. source |
| Zeplin D Zeplin, Inc. | get_screen and component tools give specs and assets, with includeVariants false and targetLayerName to cut size; the README warns large screens can overflow context. Scoping is by layer name, not node id, and acc-27's crash on untyped annotations was fixed in July. source |
Doesn't do this job 7
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| ArtCraft F Storyteller | An image and video generation app; the README describes no API, MCP server or CLI an outside agent can call, and no UI frames. source |
| Framer Server API C Framer B.V. | Framer builds and hosts its own sites; nothing exports a frame into another codebase's framework. source |
| Impeccable BB Renaissance Geek, Inc. | No design-file input; it critiques and refines interface code. source |
| Melius C Melius AI, Inc. | Melius is a generation canvas of media nodes; it has no UI frames to turn into code. source |
| Miro API + MCP B RealtimeBoard Inc. dba Miro | Miro boards hold no UI frames with design data; there is no design-to-code tool. source |
| Presentations.AI MCP server E Presentations.AI | Generates slide decks; no design file to read into code. source |
| UXKIN C UXKIN (no legal entity named) | Searches a library of other products' screens and design systems; it reads no frame from the person's design tool. source |
What stops agents finishing
- remote MCP accepts only allow-listed clients
- read seat capped at a handful of MCP calls a month
- asset download reports success without writing the file
Around it
Before: chat.user-pastes-link. After: code.implement-ui, browser.screenshot-compare.
Where this comes from
Weight 27 of 100 in Design workspaces & canvases: 12 of 42 accounts (Figma MCP, Framelink, Zeplin, Magic Patterns); the job most HN users describe doing
The same as data: /api/v1/use-cases/design.frame-to-code.json, or the MCP tool get_use_case with id design.frame-to-code. Other jobs in this category: /api/v1/niches?category=design.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.