Figma's MCP Opens to 20 Partners, CEO Admits Flaws [2026]

Figma spent October 15 telling developers it wants to be the connective tissue between design and code. Five days later, its own chief executive told reporters the standard holding that vision together is not built right yet. At Config India, the company’s first conference on the subcontinent, Figma announced 20 new partner integrations for its Model Context Protocol (MCP) support, pulling in names like JetBrains’ Junie, Notion, Linear, Mistral, and the open source coding agent OpenCode. On October 20, co-founder and CEO Dylan Field told TechCrunch that MCP “is not a perfect protocol” and has “tons of room for improvement,” adding that the standard is “advantaged more to frontier labs.”

The timing matters. MCP, the open protocol Anthropic introduced in November 2024 to let AI agents talk to outside tools and data sources, has become the default plumbing for a fast-growing list of coding agents and design platforms. Figma’s expansion shows how deeply that plumbing has wired itself into everyday developer workflows, from OpenCode sessions to JetBrains’ agent lineup. Field’s public skepticism, delivered the same week Figma leaned harder into the protocol, is a reminder that even the companies building on top of MCP are not convinced it is finished.

Google · Preferred Sources

Don't miss new tech stories on Google

Add TrendinTech once in the Google app and our stories appear in your news suggestions.

Add Now

What Figma Actually Announced on October 15

Config India was billed by Figma as its largest international conference to date, and the company used the stage to push several products out of their testing phases at once. Specs, a markdown-based container for product context that can move information into and out of Figma through MCP and GitHub, launched in open beta on web, desktop, and mobile for paid plans. Code layers, which lets teams generate and edit working code directly on the Figma canvas, moved into open beta across all plans. Figma said creating and editing inside code layers will stay free on paid plans through November 4, 2026, after which it will draw on the platform’s AI credit system.

Two more pieces of the release wave extend Figma’s reach into content and governance. Payload, the headless content-management platform Figma absorbed in 2025, became available across all plans with staged account rollout. Guidelines entered open beta on paid plans, and a feature called code sources entered closed beta, also on paid tiers. Earlier in the month, on October 6, Figma’s AI design agent exited beta entirely, a separate but related step in the company’s push to make its canvas workable by both humans and autonomous agents.

The MCP piece tying all of this together got a quieter but important technical upgrade. Figma said its MCP server now runs with fewer errors and lower token usage, and added support for generating code layers, editing Specs, creating images through a generate_image call, converting raw HTML into editable Figma layers via html_to_figma, and running published workflows built in Figma’s Weave tool. None of that is exotic by AI-tooling standards in 2026, but it is exactly the kind of plumbing that decides whether an agent-driven workflow feels fast or clunky.

The 20 New MCP Partners

Figma’s MCP Catalog, its directory of tools that can plug into the Figma canvas through the protocol, grew substantially with the Config India announcement. The company named Amp, ClickUp, Dia, Junie by JetBrains, Linear, Meta Muse, Mistral, Notion, OpenCode, Sarvam Agents, Agentforce Vibes, Amazon Quick, Amplitude Wave, Cline, Delta, Dust, Glass, Gradial, Writer, Zed, and Microsoft 365 Copilot as part of the expansion. That is a longer list than the “20” label Figma used, and the company has not clarified the discrepancy, but the roster gives a clear picture of who wants a seat at the table: coding agents, project-management tools, browsers, and at least one rival design-adjacent AI system in Meta Muse.

JetBrains’ inclusion is the one worth watching most closely for developers. JetBrains confirmed on October 15 and 16 that Junie, its AI coding agent, is now an official partner in Figma’s MCP Catalog, meaning design context captured in Figma, including the new Specs format, can flow directly into a JetBrains IDE session without a developer manually copying screenshots or writing a brief by hand. For teams already running JetBrains Air alongside Figma, that closes a loop that used to require a human translator sitting between design and code.

Figma also used the India event to announce a localized MCP partnership with Sarvam, the Indian AI company, connecting Figma workflows to Sarvam’s Indus model. The company framed that alongside its existing MCP relationships with Claude Code, ChatGPT, Cursor, GitHub Copilot, Xcode, and VS Code, arguing that design context should be portable across whichever coding agent a team happens to use. According to Figma, the India-focused research behind the announcement drew on 507 survey responses from designers, developers, and product managers in the country, part of a broader pool of more than 1,800 respondents surveyed globally between August 17 and September 16, 2026.

Category	Named MCP Partners
Coding agents and IDEs	Junie by JetBrains, OpenCode, Cline, Zed, Microsoft 365 Copilot
Browsers and productivity	Dia, Notion, ClickUp, Linear, Writer
AI model and agent platforms	Meta Muse, Mistral, Amp, Dust, Glass
Enterprise and commerce tools	Agentforce Vibes, Amazon Quick, Amplitude Wave, Sarvam Agents, Gradial, Delta

Dylan Field’s Case Against MCP

What makes the October 20 TechCrunch interview notable is that Field is not an outside critic. Figma is actively expanding its MCP footprint while its own CEO argues the protocol has structural problems. “I’m really excited about what we’re doing in terms of opening up our ecosystem. I also think that MCP is not a perfect protocol. And I think that it is something that has tons of room for improvement,” Field told TechCrunch.

Field’s specific complaint centers on who benefits most from the current design of MCP. “It is advantaged more to frontier labs,” he said, pointing to the imbalance between companies like Anthropic, which built and governs the protocol, and platform companies like Figma that are expected to open their data to any MCP client that asks. He went further on the asymmetry point: “I find it interesting that Anthropic (creators of MCP standard) do not expose artifacts / other cloud resources created via MCP,” a jab at what he sees as a one-way flow of context and access.

Field also pushed back on how the industry talks about open standards in general. “The debate is framed as open vs closed. Everyone always prefers open… obviously. And meanwhile issues of asymmetric context sharing are ignored,” he said, arguing that the open-versus-closed framing lets bigger players avoid scrutiny of how much access they actually grant in return. Figma’s own response, according to Field, is to limit how much of its data it exposes through MCP precisely because wider access raises the risk of abuse. Even with those reservations, Field described Figma’s long-term direction in expansive terms, telling Moneycontrol that “the next evolution is a full-stack creative canvas where teams and agents can build together, from idea to ship, in a visual-first way.”

Why MCP Became the Default Plumbing

MCP’s basic pitch has not changed since Anthropic introduced it in November 2024: give AI agents a common way to read and act on external tools and data, instead of forcing every vendor to write a custom integration for every agent. Two years on, that pitch has pulled in an unusually broad coalition of browsers, IDEs, project trackers, and now design tools, which is exactly the dynamic Figma’s announcement reflects. The protocol’s maintainers have kept iterating on the underlying spec rather than freezing it. A revision published on July 28, 2026 reworked MCP toward stateless, HTTP-native operation and removed the older protocol-level session and initialization handshake, a change the maintainers described as a redesign rather than evidence the protocol was stalling.

Lead maintainers David Soria Parra and Den Delimarsky followed that with a roadmap published August 22, 2026 that prioritized HTTP-native transport, stateless operation, identity and authorization controls, and progressive discovery mechanisms, alongside continued extension development. That roadmap is effectively an acknowledgment of the same gap Field is pointing at: MCP’s authorization and identity model is still being built out, which is part of why a platform holding as much proprietary design data as Figma is cautious about how wide it opens the door.

That caution is not unique to Figma. Across the fast-moving market for AI coding agents, vendors have had to balance giving agents broad tool access against the risk of a poorly scoped MCP server leaking credentials, source code, or private project data. The protocol’s growth has outpaced the tooling needed to audit every server a team might connect, which is one reason governance questions like Field’s keep resurfacing even as adoption climbs.

Figma’s Broader AI Bet: Specs, Code Layers, and Payload

The MCP expansion is one piece of a larger strategy Figma has been building out all year. Specs is the clearest expression of that strategy: rather than treating a product brief as a static document, Figma wants it to be a living, markdown-based artifact that an engineer’s coding agent can query directly through MCP. Code layers takes the same logic in the other direction, letting a generated or hand-written code snippet live on the canvas next to the visual design it represents, so a designer and an engineer are looking at the same object instead of two disconnected files.

Payload’s integration fills a third gap: content and data. Since joining Figma in 2025, Payload has been positioned as the bridge between a visual design and the structured data that eventually powers a real website or app, and its rollout across all Figma plans this month signals Figma wants that bridge available by default rather than as an add-on. Taken together, Specs, code layers, and Payload describe a company trying to make the design canvas a place where both humans and coding agents can work without constantly exporting, re-importing, or manually re-typing context, which is also exactly the workflow MCP is supposed to make possible.

Figma has leaned on AI internally as well as externally. The company said in a security blog post dated July 29, 2026 that an internal AI agent used for alert triage, forensic investigation, data-lake queries, and code fixes had cut alert-resolution time by 71%, and it announced ISO 42001 certification, an AI management-system standard, on July 1, 2026. Those moves suggest Figma is trying to demonstrate it can govern AI responsibly even as it hands more of its own platform over to third-party agents through MCP.

JetBrains Junie’s Side of the Story

JetBrains has had an active month of its own, which is part of why its MCP Catalog listing lands at a useful moment for developers deciding which coding agent to standardize on. Junie left beta and became generally available on September 17, 2026, and JetBrains folded it into a broader agent-oriented platform called JetBrains Air, announced September 22, 2026. On October 2, JetBrains introduced Junie Lite, a free tier built for routine work and currently running on the Qwen 3.6 27B model, with frontier models reserved for harder tasks.

JetBrains also pushed Junie further toward local, on-device execution. Junie Local, which runs entirely on a developer’s own machine with no token charges, expanded to stable Windows support on NVIDIA’s RTX Spark hardware on October 16, 2026, requiring an Ampere-generation or newer RTX GPU with at least 24 GB of VRAM, such as an RTX 3090, 4090, or 32 GB RTX 5090. That local option matters for the MCP conversation Field is raising: a developer running Junie Local against a Figma MCP server is sending design context to a model running on their own hardware rather than a frontier lab’s cloud, which is a meaningfully different trust model than routing everything through a hosted agent.

Competitive Field: Design Tools Racing Toward Agentic Workflows

Figma’s MCP push does not happen in a vacuum. The broader AI coding tools market has been consolidating around a handful of distribution strategies: hosted agents bundled into IDEs, standalone CLI agents, and now browser- and canvas-native agents that sit closer to where non-engineers work. Claude Code’s plugin ecosystem and Cursor’s environments API represent one branch of that race, built around developers who live inside a code editor all day. Figma’s bet is different: it is trying to make the design canvas itself an entry point into that same agentic workflow, so a product manager or designer who never opens an IDE can still hand off structured context to a coding agent.

That makes Meta Muse’s presence on Figma’s partner list notable. Meta Muse, which has been positioned as a rival to established AI coding tools, joining Figma’s MCP Catalog alongside JetBrains, OpenCode, and Microsoft 365 Copilot suggests none of the major AI labs or platform vendors want to be left out of design-to-code workflows, even when they compete head-to-head elsewhere. It also means Figma’s MCP Catalog now functions less like a curated partner program and more like a neutral meeting ground where otherwise competing coding agents all plug into the same design data.

Whether that neutrality holds is an open question. Field’s criticism that MCP is “advantaged more to frontier labs” implies he sees an uneven playing field even within that neutral-looking catalog: the labs that helped define the protocol’s rules have more influence over how it evolves than platform companies that simply adopt it. That tension, between open standards and who actually shapes them, is likely to keep surfacing as more companies build critical workflows on top of MCP rather than around it.

Historical Context: From Static Handoffs to Agentic Pipelines

It is worth remembering how recently the design-to-code handoff was a manual, often painful process. For most of the last decade, a typical workflow involved a designer finishing a mockup in Figma, exporting assets, and writing (or verbally explaining) a specification that an engineer then interpreted by hand, often losing fidelity along the way. Figma’s Dev Mode, introduced years before the current AI wave, was an early attempt to narrow that gap by giving engineers a code-aware view of a design file. MCP support, and now Specs and code layers, represent the next step: instead of a human engineer interpreting a static design, an AI agent can query the design file directly and generate working code against it.

That shift mirrors what has happened across the broader developer-tools market, where AI coding agents have moved from novelty to default infrastructure in roughly two years. MCP’s own history compresses that same arc: a protocol introduced in November 2024 has already gone through a major mid-2026 spec rewrite and is now central enough to developer workflows that a CEO as prominent as Field feels compelled to publicly critique its governance rather than quietly working around it.

Security and Governance Concerns Developers Should Watch

Field’s comments double as a public airing of concerns that security teams have been raising about MCP privately for a while. The core risk is that MCP servers do not just expose read-only information, they can expose actions, which means a developer approving a tool call is effectively granting an agent permission to do something, not just look at something. A poorly scoped or poorly audited MCP server can leak credentials, source files, or private project data to whatever agent is connected to it. Prompt injection is a related risk: instructions buried inside a retrieved document or a piece of web content can manipulate an agent into taking an action the developer never intended.

Figma’s own stated response, limiting how much data it exposes through MCP specifically to reduce abuse risk, is a practical acknowledgment of that threat model. It also illustrates the tradeoff every platform adopting MCP now faces: the more useful an MCP server is to outside agents, the more surface area it creates for something to go wrong. The protocol’s own roadmap work on identity and authorization, flagged by maintainers in the August 2026 update, is a direct attempt to close that gap, but it is still in progress rather than finished, which lines up with Field’s point that the standard has “tons of room for improvement.”

Market Impact: What This Means for Developer Tool Vendors

For vendors building coding tools, the practical takeaway from Figma’s announcement is that MCP support is no longer optional if you want to be part of a design-to-code pipeline. Being named in Figma’s MCP Catalog is becoming a visible signal of relevance, the same way appearing in a major cloud provider’s marketplace once was. That creates pressure on any coding agent, IDE, or project-management tool not already on the list to either build MCP support or risk being excluded from workflows that increasingly start inside a design file rather than a code editor.

It also raises the stakes for the protocol’s own governance. If MCP keeps absorbing more of the connective tissue between design, project management, and coding agents, the complaints Field raised, about frontier labs having disproportionate influence and about asymmetric data sharing, stop being abstract and start shaping who controls access to a growing share of enterprise software development. Figma is big enough that its criticism carries weight, but it is also dependent enough on the protocol’s continued growth that it has little choice but to keep building on it anyway.

Milestone	Date	Detail
MCP introduced	November 2024	Anthropic releases MCP as an open protocol for connecting AI agents to external tools and data
Figma Agent exits beta	October 6, 2026	Figma’s AI design agent becomes generally available with improved output quality
MCP spec rewrite	July 28, 2026	Protocol moves toward stateless, HTTP-native operation; removes legacy session handshake
Junie reaches general availability	September 17, 2026	JetBrains’ coding agent exits beta
JetBrains Air announced	September 22, 2026	JetBrains frames Junie as part of a broader agent-oriented platform
MCP roadmap published	August 22, 2026	Maintainers prioritize identity, authorization, and stateless transport
Junie Lite launches	October 2, 2026	Free Junie tier running on Qwen 3.6 27B
Config India / Build by Design	October 15, 2026	Figma announces Specs, code layers, Payload, and 20 new MCP partners
Junie Local reaches RTX Spark	October 16, 2026	Stable Windows support for 24 GB+ Ampere-or-newer RTX GPUs
Field criticizes MCP governance	October 20, 2026	Figma CEO tells TechCrunch MCP is “not a perfect protocol”

What Comes Next: Five Predictions

Expect more design and product-management platforms beyond Figma to publish their own MCP catalogs rather than building isolated, one-off AI features, since Figma’s approach of opening a managed catalog looks easier to maintain than bespoke integrations with every coding agent.
Expect MCP’s governance debate to intensify rather than fade, with more platform CEOs willing to publicly criticize the protocol’s authorization model even as they keep shipping MCP support, following the pattern Field set this week.
Expect JetBrains to keep pushing Junie toward on-device execution, since local models sidestep some of the data-exposure concerns Field raised about routing design context through frontier-lab infrastructure.
Expect the gap between Figma’s “20 partners” framing and the longer list of names it actually published to prompt scrutiny over how design-tool vendors count and market MCP integrations, especially as more companies claim MCP support without clarifying what tier of integration that means.
Expect identity and authorization, the exact gap flagged in the MCP maintainers’ August roadmap, to be the next major spec-level fight, since it is the technical area most directly tied to the abuse-risk concerns both Figma and security researchers have raised.

Frequently Asked Questions

What is the Model Context Protocol (MCP)?

MCP is an open protocol, originally released by Anthropic in November 2024, that gives AI agents a standard way to connect to external tools, applications, and data sources instead of requiring custom integrations for every pairing of agent and tool.

What did Figma announce about MCP on October 15, 2026?

At its Config India conference, Figma announced 20 new MCP partner integrations, including JetBrains’ Junie, OpenCode, Linear, Notion, and Mistral, alongside new products Specs, code layers, Payload, Guidelines, and code sources.

Why did Figma’s CEO criticize MCP?

Dylan Field told TechCrunch on October 20, 2026 that MCP is “not a perfect protocol” and is “advantaged more to frontier labs,” arguing that the standard creates an asymmetry where platforms like Figma are expected to expose data broadly while large AI labs do not reciprocate.

Is JetBrains Junie part of Figma’s MCP Catalog?

Yes. JetBrains confirmed on October 15 and 16, 2026 that Junie, its AI coding agent, is an official partner in Figma’s MCP Catalog, allowing design context created in Figma to flow into a JetBrains coding session.

What is Junie Local and why does it matter for MCP?

Junie Local is JetBrains’ on-device version of its coding agent. It gained stable Windows support for NVIDIA RTX Spark and other Ampere-or-newer RTX GPUs with at least 24 GB of VRAM on October 16, 2026. Because it runs locally, it processes MCP-delivered context on the developer’s own machine rather than sending it to a hosted model.

Is Figma’s code layers feature free to use?

Figma said creating and editing within code layers, which entered open beta on all plans on October 15, 2026, is free on paid plans through November 4, 2026. After that date it draws on Figma’s AI credit system.

What is Figma Specs?

Specs is a markdown-based container Figma introduced on October 15, 2026 for turning product context into detailed specifications that can be passed to coding agents through Figma’s MCP integration. It launched in open beta on web, desktop, and mobile for paid plans.

Does MCP competition exist from other protocols?

Google has developed Agent2Agent (A2A), a protocol focused on letting AI agents communicate and coordinate tasks with each other, which is generally viewed as complementary to MCP’s focus on connecting agents to tools and data rather than a direct replacement for it.

Figma Opens MCP to 20 AI Partners as CEO Calls It Imperfect [2026]