When should you use it?
- You design in Figma and want a quick coded version
- Screenshots fed to AI tools gave weak results
- You can open a project folder in an AI editor
Not for: Production builds without developer review; Designs relying on paid fonts you have not licensed for the web
What do you need, and what do you get?
You need
- A Figma file with Dev Mode access
- An AI code editor with the Figma bridge enabled
- A local repository
You get
- HTML and CSS for the selected frame
- A list of gaps, such as fonts, to fix
How does it work, step by step?
- Set up a repoCreate a local folder under version control and open it in the editor.
- Connect the design bridgeEnable Figma's MCP connection in the editor.
- Select and linkIn Dev Mode, select the frame and copy its link.
- Prompt simplyAsk for semantic HTML and CSS matching the frame.
- Compare and tweakCheck fonts, spacing and states against the design and iterate.
The prompt
Our own prompt, inspired by Phil Sayer and written to work in Custom agent. Values in lime are placeholders.
Here is a Figma frame link: [link]. Using the design data, build this section as semantic, accessible HTML with a separate CSS file. Match colors, spacing, layout and hover states exactly. Use CSS variables for colors and sizes. If a font is not available as a web font, use the closest fallback and tell me which. List any part of the design you could not reproduce.
Inspired by Phil Sayer: Getting remarkable results quickly with Figma MCP and Cursor. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Custom fonts may not transfer; choose a licensed web font or self-host it.
- Generated markup can be div-heavy; ask for accessible, semantic structure.
Our take
Phil Sayer's comparison makes a simple point clearly: an AI given a picture guesses, while an AI given the design's actual properties can build from them. Pulling colors, spacing and component states straight from Figma is why one prompt captured even hover effects that a screenshot never shows. For designers, that lowers the cost of seeing a layout live in the browser, which is where problems with responsive behavior and real content appear. The missing font is a reminder that design files reference assets the code environment may not have. Licensed typefaces like Rotis need web licenses and font files before they will render, so plan for that rather than accepting a fallback. Ask from the first prompt for semantic, accessible markup, real headings, buttons and labels, and for responsive breakpoints, because retrofitting those later costs more than generating them correctly. Compare the result side by side with the frame at several widths, and tidy the design file's layer naming first, since clear names produce clearer class names and code.
Where the time goes back: The first build of a page goes from hours to one prompt, leaving only refinements.
FAQ
Why not screenshots?
What did it miss?
Source and credit
Inspired by Phil Sayer. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- Getting remarkable results quickly with Figma MCP and Cursor
- Creator
- Phil Sayer
- Published on
- designphilosophy.substack.com (Substack)
- Original date
- Aug 5, 2025
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Custom agent
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.