How We Automated Design Audits with Figma MCP and Jira

So developers and QAs never have to ask “where’s the error state?” again.

❓ The Problem

We get the same Slack message time to time: “Hey, where’s the error state for this screen?” or “What happens at the end of this flow?”

Every time we get this message, there’s a context switch for the designer, a delay for the developer, and a small but real loss of momentum. The root cause isn’t carelessness. It’s that complex files grow fast, and there’s no systematic check in place before handoff.

⚒️ What We Built

A Figma MCP-powered design audit automation that runs directly inside Jira. A designer adds a Figma link to a task, triggers the automation, and gets back a structured audit report as a comment — no context switching, no separate tool.

The audit checks the things that most commonly cause friction during development: possible flows, UI states, error handling, and edge cases.

Photo by Alexander Grigoryev on Unsplash

🧩 How the Output Is Structured

Summary: A quick overview of overall design coverage so anyone reading the task knows where things stand.

Categorized Findings (High / Medium / Low) in One Table with Suggestions: Each audit category has its own section with findings ranked by severity. Without prioritization, a list of 30 findings is overwhelming. With it, we know exactly where to start. Also, it suggests us how to resolve the issue.

Figma Node IDs for critical issues: This turned out to be the most valuable part. For every high-severity finding, the comment includes a direct Figma link. Click it, and you land exactly on the frame in question. A finding that says “the checkout screen is missing an error state” is helpful. One that takes you there with one click is something people actually act on.

🦾 How We Built It

1. Design coverage checklist: Before writing any prompt, we defined what a well-designed file actually needs, broken into categories with mandatory and optional items. This surfaced useful team disagreements before they got encoded into automation.

2. Prompt generation with PromptCowboy: We used the checklist to generate a structured prompt. Most of the work went into output format — we needed findings categorized, ranked, and readable inside a Jira comment.

3. Testing with Figma MCP via Antigravity: We tested against files of different sizes and complexity. Each run revealed something: too verbose here, missing a category there, format breaking on large files. We kept notes and revised.

4. Refinement until stable: The target was consistent, reliable output regardless of who ran it or what file it pointed at. This took more iterations than expected.

5. Jira automation setup: Once the prompt was stable, our QA team wired it into Jira. The full flow: add Figma link → trigger automation → audit appears as a comment.

⏭️ What’s Next

We’re turning this into a Claude skill so anyone can paste a Figma link directly into Claude and get the same structured audit — without needing to go through Jira at all. The prompt and checklist are already there. The packaging is what’s left.

figmamcp #jiraautomation #designaudit #handoverquality