Figma Adds Codex Days After Claude Code - MCP Wins

Figma just made itself the Switzerland of AI coding tools - and MCP is the passport.

On February 26, the design platform announced its integration with OpenAI's Codex, exactly nine days after launching the same capability with Anthropic's Claude Code. Both integrations run through Figma's MCP server, the same protocol, the same bidirectional pipeline. Designers push layouts into code. Engineers push running UIs back into editable Figma frames. The AI model doing the work is now interchangeable.

This isn't a story about two partnerships. It's a story about a protocol becoming infrastructure.

TL;DR

Figma integrated OpenAI Codex on Feb 26, nine days after integrating Claude Code on Feb 17
Both use the same Figma MCP server for bidirectional design-to-code and code-to-design workflows
Over 1 million weekly Codex users and growing Claude Code adoption can now work directly with Figma designs
The real winner is MCP - one protocol, multiple AI backends, no vendor lock-in

How It Works

Design to Code

The Figma MCP server exposes a get_design_context tool that extracts structured design data - layouts, styles, component metadata, auto-layout rules, and design tokens - directly from Figma files. When a developer pastes a Figma frame URL into Codex or Claude Code, the AI does not guess how a button should be styled from a screenshot. It reads the actual variable values, layer hierarchies, and spacing rules defined by the designer.

This is a meaningful shift from the old workflow of inspecting layers, copying hex codes, and arguing about padding. The AI agent reads the source of truth and produces code that respects it.

Code to Design

The generate_figma_design tool runs in the opposite direction. A developer builds a UI in code, runs it locally, and Codex captures the live browser state - converting it into fully editable Figma frames in seconds. Designers can then explore alternatives, adjust spacing, swap components, and confirm assumptions without touching a line of code.

"With this integration, teams can build on their best ideas - not just their first idea - by combining the best of code with the creativity, collaboration, and craft that comes with Figma's infinite canvas," said Loredan Crisan, Figma's Chief Design Officer.

UX design prototyping workspace with wireframes and design tools The design-to-code gap has been one of the most persistent friction points in product development. Figma's MCP integrations aim to close it completely.

The Comparison

Feature	Codex Integration	Claude Code Integration
Launch Date	February 26, 2026	February 17, 2026
Protocol	Figma MCP Server	Figma MCP Server
Design to Code	Yes (`get_design_context`)	Yes (`get_design_context`)
Code to Design	Yes (`generate_figma_design`)	Yes (`generate_figma_design`)
FigJam Support	Yes	Yes
Figma Make Support	Yes	Yes
Weekly Active Users	1M+ (Codex)	Not disclosed
Platform	macOS desktop app	Terminal CLI

The feature set is identical because the Figma MCP server is the same regardless of which AI model connects to it. This is exactly what MCP was designed to do - decouple the tool from the model.

Why MCP Matters Here

The Model Context Protocol, originally developed by Anthropic and now an open standard, provides a standardized interface for AI agents to interact with external data sources. Figma's MCP server doesn't care whether it is talking to Codex, Claude Code, Cursor, Windsurf, or any other MCP-compatible client.

This is the point. Figma built one integration and got access to every major AI coding tool at once. Developers pick the model they prefer. Designers don't need to care which one their engineers chose. The design data flows the same way either direction.

Code editor with software development workflow on screen Codex and Claude Code both connect to Figma through the same MCP server, making the AI model an interchangeable component in the design-to-code pipeline.

The Competitive Angle

The timing is deliberate. Figma isn't picking sides between OpenAI and Anthropic - it's making itself vital to both ecosystems.

Alexander Embiricos, Codex product lead at OpenAI, framed the integration as an expansion of Codex's audience: "The integration makes Codex powerful for a much broader range of builders and businesses because it doesn't assume you're 'a designer' or 'an engineer' first."

In parallel, Figma CEO Dylan Field told CNBC on February 17 that the company is "embracing AI" as a core part of its product strategy. Field has spoken publicly about what he calls the "software reckoning" - a moment where AI tools fundamentally change how products get built. Figma's move to integrate with both major AI coding tools simultaneously suggests the company sees itself as the design layer that all AI coding agents will flow through.

For context, Figma was among the first companies to launch an app in ChatGPT back in October 2025. The OpenAI relationship is not new. But shipping the Codex integration just nine days after Claude Code signals urgency. In a market where every AI coding tool is fighting for developer attention, Figma is playing kingmaker - whichever model wins, Figma wins too.

What It Does Not Tell You

The Gap Between Demo and Production

The generate_figma_design tool captures visual UI state - what the browser renders on screen. It doesn't capture interactive behavior, state management, or the logic behind a component. A captured button looks right in Figma but has no click handler, no loading state, no error condition. Round-trip fidelity has limits.

MCP Is Still Beta

Figma's MCP server is in beta. Users have already filed issues on GitHub about generate_figma_design not appearing in Codex's tool catalog. The integration works when it works, but the developer experience isn't yet polished.

The Missing Players

Figma's MCP server technically supports any MCP-compatible client, but the company has only officially announced partnerships with OpenAI and Anthropic. Cursor and Windsurf connect through community-built wrappers. Google's Gemini has no official Figma MCP integration. The "universal bridge" narrative is aspirational - not yet fully realized.

Team collaborating on software design at computer workstations Figma's strategy positions it as the design layer that bridges every AI coding agent - regardless of which model powers them.

Design Tokens Are Not Design Taste

An AI agent can read that a button has 16px padding, a 4px border radius, and uses the color token primary-600. It can't tell you whether that button belongs in a modal or a page header. Design decisions still require a human who understands the product, the user, and the context. The tools close the handoff gap - they don't close the judgment gap.

Figma's dual integration bet is a sharp strategic move. By routing both Codex and Claude Code through the same MCP server, the company has turned a protocol into a moat. Designers get a single workflow regardless of which AI model their engineering team uses. Developers get design context that's structured, accurate, and machine-readable. And Figma gets to be the vital middle layer in a market where every other tool is fighting to own the edges. The protocol won. Figma made sure it was standing in the right place when it did.

Sources: