Figma's Dylan Field Says Model Context Protocol Is Not Perfect
Figma adds 20 MCP partners at its first Config conference in India while Field questions the open standard's design and its trade-offs
By AXY AI Desk ·
Key points
- Figma co-founder Dylan Field said the Model Context Protocol is 'not a perfect protocol' and has 'tons of room for improvement.'
- Figma announced 20 new MCP partners at its first Config conference, held in India.
- Field argued that MCP is advantaged more to frontier AI labs and raised concerns about asymmetric context sharing.
- Figma says exposing more data through MCP increases the chance of abuse, so it must weigh those risks while supporting open standards.
Figma co-founder Dylan Field has said that the Model Context Protocol (MCP) is 'not a perfect protocol,' even as the design software company expands its support for the open standard. Speaking at a media briefing, Field said MCP 'has tons of room for improvement' and 'is advantaged more to frontier labs.' His remarks came as Figma announced 20 new MCP partners at its first Config conference, held in India. The partners include Amp, ClickUp, Dia, Junie by JetBrains, Linear, Meta Muse, Mistral, Notion, OpenCode and Sarvam Agents, according to TechCrunch.
What the Model Context Protocol does
The Model Context Protocol is an open-source standard that defines how AI agents connect to external data sources and tools. An AI agent is software that can carry out tasks on a user's behalf, such as finding information, reading files or updating records in another application. For those agents to be useful, they need a way to reach the systems where that information lives.
Without a shared standard, each AI tool would need a custom connection to each software product it wanted to use, which multiplies the engineering work for everyone involved. MCP works more like a common plug design: if a product and an AI agent both follow the same rules, they can connect without a bespoke integration for every pair. The standard was created by Anthropic, the AI company behind the Claude assistant, as Field's own post on X makes clear when it refers to 'Anthropic (creators of MCP standard).'
The practical appeal is easy to see. A designer, product manager or developer could, in principle, let an AI assistant pull information from a design file or project board and act on it, without the software vendor building a separate system for that assistant. The same convenience also means that the rules governing what an agent can see and do become important, because a single connection can expose a large amount of material at once.
How the debate reached Figma's Config event
Field's concerns about MCP did not begin with the Config announcement. TechCrunch reports that days before the event, one of MCP's co-creators complained about Figma not supporting Pi, an open-source tool. In responding, Field turned the criticism around by pointing to how Anthropic itself handles resources created through MCP.
In a post on X, Field wrote: 'I find it interesting that Anthropic (creators of MCP standard) do not expose artifacts / other cloud resources created via MCP.' He went on to argue that the public conversation is framed too simply. 'The debate is framed as open vs closed,' he wrote. 'Everyone always prefers open… obviously. And meanwhile issues of asymmetric context sharing are ignored.'
Asymmetric context sharing describes a situation in which information flows unevenly between parties. One side may be able to access, store or reuse material that the other side cannot, or the rules for sharing may favour one party. In the context of AI agents, this could mean that a large AI developer benefits from the context a platform shares, while the platform and its users receive less in return. Field's argument is that the more important question is not simply whether a standard is open, but who gets to see what, and on what terms. TechCrunch's report does not include a response from Anthropic to Field's remarks.
Figma's expanding support
Against that backdrop, Figma's announcement at Config centred on widening the set of companies that can connect with Figma through MCP. The 20 partners are a varied group, and their inclusion suggests that Figma wants design files and related data to flow into many different workflows rather than stay inside its own application. Among the names are Notion and Linear, which are widely used for documents and project tracking, as well as Mistral and Sarvam Agents, which are AI-focused companies or products.
Figma has said it plans to support as many standards as possible without exposing all of its data. That sentence captures the central tension in the story. Joining an open standard can make a product more useful to customers, because they can use it with the tools they already rely on. But each additional connection is also another route through which information can travel. The partner list shows that Figma is choosing to take part in the standard, while its leaders have made clear that participation has limits.
For design and product teams, the stakes are concrete. Design files can reveal unreleased product plans, internal organisation and commercial strategy. Companies therefore pay close attention to who can read that material and whether it can be sent to outside systems. A decision by a major design platform about how much data to expose through a common standard can affect how other software companies approach the same question.
Field's argument at the briefing
At a media briefing, Field explained his position in more detail. He said that while Figma wants to be open, he does not believe MCP is a perfect protocol. 'I'm really excited about what we're doing in terms of opening up our ecosystem,' he said. 'I also think that MCP is not a perfect protocol. And I think that it is something that has tons of room for improvement.'
His most pointed criticism concerned who benefits from the standard. 'It is advantaged more to frontier labs,' he said. Frontier labs are the leading AI developers building the most capable general-purpose models, and they generally have large computing budgets and research teams. A concern that a standard favours these organisations is a concern that its design may give the largest AI companies advantages that smaller developers, startups and software platforms do not share.
Field also said he understands people online who want a protocol they can use whenever they want to work with their Figma data. 'I want them to use it, too,' he said. That statement signals that Figma is not rejecting MCP outright. Instead, the company is trying to balance user access with the risks that come from exposing more data to outside tools. The position is less a simple yes or no than an attempt to define how far openness should go.
The security trade-off
TechCrunch reports that Figma's chief executive explained why the company does not want to be fully open. According to the report, the more data the company exposes through MCP, the higher the chance of abuse. The executive also clarified that most users do not intend to use MCP or Figma data in a malicious way. Even so, the company has to consider those potential risks.
This is a familiar dilemma for platforms that hold valuable business information. Access controls are designed to let legitimate users do their work while keeping sensitive material away from people and systems that should not see it. Connecting a platform to an AI agent can save time and open new uses, but it also widens the number of tools, accounts and automated processes that could reach the data. Security specialists often describe this as a larger attack surface, meaning more points where something can go wrong, whether through an error, a compromised account or deliberate misuse.
The source material does not describe the specific safeguards Figma plans to use, such as limits on which data types can be shared or controls over how agents act on that data. However, the executive's remarks suggest that the company sees the answer as a matter of managing risk rather than choosing between full openness and complete closure.
What to watch next
Several questions remain open after the announcement. The first is whether the Model Context Protocol will change in response to the concerns Field raised about asymmetry and about advantages for large AI labs. Open standards typically evolve through contributions from the people who use them, so the debate could influence future versions, although the source material does not indicate that any specific change is planned.
The second question is whether other software companies will follow Figma's approach of joining the standard while setting clear limits on what they expose. If many platforms take that middle path, the practical rules for AI agents may come to depend less on the standard itself and more on the access settings each company chooses.
The third is whether Anthropic responds publicly to Field's question about why it does not expose certain artifacts and cloud resources created through MCP. Any answer would help clarify whether the disagreement is about technical design, company policy or both.
For readers, the story is a useful case study in how open standards are being tested as AI agents become more common. Openness can expand what people can do with their own data and reduce the cost of connecting tools. At the same time, it raises questions about who benefits from shared context, who controls access and how misuse is prevented. Figma's position, which supports the standard while acknowledging its flaws, suggests that other companies may face the same balancing act as they decide how far to open their platforms.
Questions and answers
What is the Model Context Protocol?
The Model Context Protocol (MCP) is an open-source standard that defines how AI agents connect to external data sources and tools. Anthropic created the standard, according to Field's post on X.
Why does Dylan Field think MCP is not perfect?
Field said MCP has tons of room for improvement and is advantaged more to frontier AI labs. He also raised concerns about asymmetric context sharing, arguing that the debate is framed too simply as open versus closed.
Why doesn't Figma expose all of its data through MCP?
Figma's chief executive said the more data the company exposes through MCP, the higher the chance of abuse. Although most users do not intend to misuse it, the company has to consider those risks.
Which companies joined Figma's MCP partner program?
At its first Config conference in India, Figma announced 20 new MCP partners, including Amp, ClickUp, Dia, Junie by JetBrains, Linear, Meta Muse, Mistral, Notion, OpenCode and Sarvam Agents.
Topics: Figma, Dylan Field, Model Context Protocol, Anthropic, AI agents, Config conference, Open-source standards
Sources: TechCrunch AI
https://axynewsglobal.com/article/figma-s-dylan-field-says-model-context-protocol-is-not-perfect