How AI reads Figma

figma.com Jun 24, 2026 6 min read Emerging

Short

52/ 100

AEO Level 3Emerging

The short answer

We scanned Figma and found an AEO score of 52 out of 100. Its structured data and agent discovery signals are solid, but agent interface support registers zero — AI assistants can parse Figma's metadata but cannot programmatically invoke or negotiate with its services in any standardized way.

What AI sees

An AI agent arriving at Figma's homepage finds a well-branded design platform with good surface metadata but no machine-callable interfaces waiting beneath it.

Figma presents itself coherently as a collaborative design tool: title tags, Open Graph metadata, and structured data are all in order, earning strong scores in both discovery and structured data categories. Agents can confirm the site's purpose, brand, and general content with confidence. Past that surface layer, however, the machine hits a wall. There is no Markdown variant of page content, no RFC 8288 Link headers directing agents to useful resources, no MCP server card, and no agent skills index. For a platform whose entire value proposition revolves around programmatic design workflows, APIs, and plugin ecosystems, the complete absence of agent-facing interfaces is a meaningful blind spot — one that will compound as AI-driven design tooling continues to accelerate.

Figma's homepage as an AI agent sees it: strong metadata layer, but no machine-callable interfaces or agent discovery endpoints surfaced

Figma's robots.txt carries no Content-Signal directives, leaving AI training, search, and input preferences entirely undeclared

Where it loses points

Agent interfaces is the single category collapsing Figma's overall score — every check in that dimension returned a failure.

How to fix it

Three targeted additions would deliver the sharpest AI readiness lift for a platform operating at Figma's scale and technical depth.

Publish an MCP Server Card

Goal

Expose Figma's capabilities to AI orchestration platforms through a standardized server card at /.well-known/mcp/server-card.json.

Issue

The scan found no MCP Server Card at /.well-known/mcp/server-card.json, leaving every agent-native tool blind to Figma's transport endpoints and declared capabilities.

Fix

Serve a JSON document at /.well-known/mcp/server-card.json with serverInfo (name, version), a transport endpoint pointing to Figma's MCP-compatible API surface, and a capabilities block. This single file turns Figma into a first-class citizen for AI orchestration platforms that auto-discover and invoke design operations on behalf of users.

Return Markdown for Agent Requests

Goal

Deliver a text/markdown version of homepage and documentation content when an AI agent signals it prefers Markdown via the Accept header.

Issue

Figma returns only HTML regardless of Accept: text/markdown requests, forcing agents to parse design-heavy markup to surface actionable information.

Fix

Add server-side content negotiation so requests carrying Accept: text/markdown receive a clean Markdown response with Content-Type: text/markdown, while HTML remains the default for browsers. For Figma specifically, this means stripping canvas embeds and surfacing feature descriptions, pricing, and API documentation in a format agents can directly quote and reason over without visual noise.

Publish an API Catalog

Goal

Surface Figma's REST, Webhooks, and Plugin APIs through a machine-readable catalog at /.well-known/api-catalog per RFC 9727.

Issue

No API catalog was detected at /.well-known/api-catalog, so automated agents cannot discover Figma's API surface without prior out-of-band knowledge.

Fix

Serve /.well-known/api-catalog as application/linkset+json listing each API with rel and type fields. AI agents and integration platforms will then auto-discover the full surface, directly reducing the friction for developers building AI-assisted design workflows on top of Figma's ecosystem.

Common questions

Why does Figma score 52 out of 100 when its structured data and agent discovery look healthy?

Figma's structured data (80/100) and agent discovery (85/100) reflect a strong metadata foundation. But AEO measures the full stack of AI readiness, and agent interfaces scored zero — every check in that category failed. Modern agents need more than metadata; they need callable interfaces and content they can negotiate. That single category pulls the composite score well below what the surface signals suggest.

What does an MCP Server Card actually unlock for a design platform at Figma's scale?

An MCP Server Card at /.well-known/mcp/server-card.json tells AI orchestration platforms exactly how to connect to Figma — transport protocol, available tools, authentication approach. Without it, agents building design automation pipelines rely on undocumented discovery or hardcoded assumptions. Publishing the card makes Figma's capabilities consumable by any standards-compliant AI agent automatically.

Who benefits most from Figma adding Markdown content negotiation?

Figma's large developer and designer audience increasingly works through AI coding and design assistants. When an agent requests Markdown and receives it, it can extract API parameters, feature summaries, and pricing cleanly — no HTML parsing required. That makes AI-assisted onboarding, plugin development, and integration work noticeably faster and more reliable for the teams building on top of Figma.