Figma expands MCP ecosystem while Dylan Field questions protocol limits
Figma announced 20 new Model Context Protocol (MCP) partners at its first Config conference in India, expanding support for the open-source standard used to connect AI agents with external data sources and tools. The additions include Amp, ClickUp, Dia, Junie by JetBrains, Linear, Meta Muse, Mistral, Notion, OpenCode and Sarvam Agents.
The expansion comes as Figma co-founder and CEO Dylan Field continues to voice reservations about MCP. Field said the company wants to open its ecosystem, but does not consider the protocol complete or without risks.
Support for MCP, with reservations
At a media briefing, Field said he was excited about Figma’s work to make its ecosystem more open. At the same time, he described MCP as a protocol with substantial room for improvement. In his view, the standard is advantaged more to frontier labs, even as users want a practical way to use their Figma data through a common protocol.
The comments followed an online exchange in which one of MCP’s co-creators criticised Figma for not supporting the open-source tool Pi. Field responded by questioning why Anthropic, which created MCP, does not expose artifacts and other cloud resources through MCP. He later said that Figma plans to support Pi.
Data exposure remains the central concern
Field framed the debate as more complicated than a choice between open and closed systems. He pointed to what he called asymmetric context sharing, arguing that this issue can be overlooked when openness is treated as the only goal.
Figma’s position is that making more data available through MCP can increase the opportunity for abuse. Field clarified that most people seeking to use MCP or Figma data do not intend malicious activity, but said the company must still account for that possibility when deciding how broadly to expose information and resources.
What the announcement means for teams
The 20-partner announcement signals that Figma is willing to broaden AI-agent connectivity while maintaining limits around access to design data. It also illustrates a wider implementation question for organisations adopting MCP: protocol support does not remove the need to decide which resources, permissions and context should be available to external tools.
For businesses, the practical implication is to evaluate MCP integrations as both a workflow opportunity and a data-governance decision, defining access boundaries before granting AI agents access to design systems or project information.