Figma CEO Dylan Field Calls MCP Imperfect as Figma Backs Multi-Standard AI Future

TL;DR

• Figma co-founder and CEO Dylan Field says the Model Context Protocol (MCP) is promising but not perfect, arguing it has real limits for complex, visual design workflows.
• Figma will support MCP alongside other emerging AI standards instead of betting on a single protocol, as part of a multi-standard, open-ecosystem strategy.
• Despite opening up to AI agents, Figma plans to keep user data guarded, emphasizing permission controls, context limits, and designer agency as AI moves into design tools.

The MCP Hype Meets Design Reality

Figma CEO Dylan Field is pouring some cold water on one of AI's hottest standards. While the Model Context Protocol, the open-source framework pioneered by Anthropic to connect AI assistants to external data and tools, has quickly become the default way to plug agents into everything from codebases to CRMs, Field argues it wasn't built with designers in mind.

In recent remarks, Field described MCP as useful but imperfect, particularly when applied to the deeply visual, iterative, and collaborative nature of product design. Unlike text-based tasks where passing files or API schemas is enough, design work involves canvas state, spatial relationships, version history, component systems, and unstated intent — context that is far harder to serialize into a prompt.

For Figma, that distinction matters. The company sits at the center of how software gets designed, and as AI agents move from generating code to generating interfaces, the fidelity of that context will determine whether they are genuinely helpful or just produce plausible-looking mockups.

Why MCP Falls Short for Design Workflows

Field's critique centers on limits, not rejection. MCP excels at giving a model standardized access to tools and data, but design workflows demand more than tool calls.

A designer working in Figma rarely executes a single linear task. They duplicate frames, test variations, leave comments, branch components, and revert decisions. Capturing that process requires persistent, high-bandwidth, bidirectional context — not just a one-off fetch of layers or assets.

There are also performance and precision issues. Streaming an entire design file's structure to a large language model can be noisy and token-heavy, while oversimplifying it risks losing the constraints that make a design system work, like auto-layout rules, variables, and accessibility requirements.

In short, Field's view is that MCP gets agents in the door, but it doesn't yet give them true design literacy.

Figma's Multi-Standard Bet

Rather than waiting for MCP to evolve or building a closed alternative, Figma says it will support multiple AI standards.

The strategy reflects a broader industry reality: no single protocol has won. Alongside MCP, tech giants and startups are pushing agent-to-agent communication frameworks, web-native agent interfaces, function-calling standards, and proprietary connectors for platforms like ChatGPT, Claude, and Microsoft Copilot.

Figma doesn't want to pick a winner prematurely and risk locking out builders or customers. By backing MCP while also integrating with other ecosystems and maintaining its own APIs and developer platform, the company hopes to meet AI models and agents wherever they emerge.

That approach also mirrors Figma's history. From plugins to Dev Mode to Figma Sites and Make, the company has grown by becoming an open platform for third-party workflows rather than dictating a single way to work.

Guarded Data in an Agentic Era

Openness, however, does not mean open data. As Figma expands AI access, Field has stressed that customer data will remain guarded.

That means granular permissions for what agents can see and do, clear boundaries between team files and model training, and controls that let organizations decide when AI can read, suggest, or edit. Figma has spent years building trust with large enterprise customers, and indiscriminate agent access to unreleased product designs would be a nonstarter.

Expect Figma to lean into scoped contexts, auditability, and human-in-the-loop approvals — allowing an agent to draft variants or clean up a design system, for example, without automatically overwriting source-of-truth files.

The stance puts Figma in line with other SaaS platforms navigating the same tension: developers want rich API access for agents, while security teams want guarantees that intellectual property won't leak through a prompt.

What It Means for the Future of AI Agents and Design Tools

Field's comments signal where design tools are headed: from AI features inside the canvas to AI agents operating across the canvas.

In the near term, that looks like agents that can reason over a Figma file from inside Claude, ChatGPT, or Cursor, generate production-ready UI from a spec, translate designs to code, and keep design systems consistent. Longer term, it points to a world where designers orchestrate fleets of specialized agents — one for research synthesis, one for wireframing, one for prototyping, one for handoff.

If MCP remains imperfect for those tasks, new layers will emerge on top of it: richer design-specific schemas, visual grounding models that actually understand pixels and layout, and tighter real-time sync between Figma's multiplayer engine and external agents.

For now, Figma's message to the AI industry is clear. It will embrace the agent era, support the standards that matter, and build bridges to every major model — but on terms that preserve craft, control, and the messy human judgment at the heart of great design.

AndroGuider Team

Articles written by the AndroGuider team. We try to make them thorough and informational while being easy to read.