The official Figma MCP server on the free Starter plan is limited to 20 calls per month, while the limit of 200 calls per day and 10 calls per minute is available only on paid Full/Dev seats — these quotas are confirmed by Figma's official documentation. This limitation can be bypassed using the open-source project figma-console-mcp by southleft: the MCP server connects to the Desktop Bridge plugin within the Figma desktop application and operates via the Plugin API, to which the official server's quotas do not apply. For the industry, this is direct pressure on the monetization of AI access to design data; for users, it is a free way to connect an agent to Figma when the monthly quota has already been exhausted.

What happened

The confirmed facts are as follows. The limits apply specifically at the level of the official integration point, the MCP server: on the Starter plan, only tool calls that read data from Figma are billed, while write tools are not affected by billing. The workaround is built on changing the API surface: the figma-console-mcp MCP server by southleft connects not to the official server, but to the Desktop Bridge plugin within the Figma desktop application, and all agent operations go through the Plugin API, to which official quotas do not apply. The project has been developing since October 2025 and is distributed under the MIT license as the npm package figma-console-mcp. The current release v1.46.0 from October 29, 2026, claims 130 tools in local mode and 105 in cloud mode, which is designed for web clients like Claude.ai. These include creating and editing frames, components, and layers, managing and synchronizing design tokens, taking screenshots, monitoring edits, and working with FigJam and Slides. The route works on macOS and Windows.

Context

The official server's restrictions are an artifact of monetization, not a property of the design data itself: the limits are built into a single integration component, the MCP server, and generous daily quotas are tied to the seat type. As long as the mandatory restriction is in place at the single entry point, its bypass is architecturally predictable: it is enough to switch the access point from the official MCP server to the Plugin API within the desktop application, and the point of restriction is shifted. This is exactly the combination — MCP server plus Desktop Bridge plugin — that figma-console-mcp implements. It is significant that the claim of a complete absence of limits for the Plugin API comes from the project author and repository: there are no official confirmations from Figma; this is an observation, not a contract. On a broader background, Figma has essentially carved out AI access to design data as a separate paid unit, while the open-source community has shown that a free route technically leads to the same data. This conflict between the paid entry point and the bypass channel is the essence of the news context.

Why this matters for the industry

Figma has effectively monetized AI access to design data by embedding limits into one of the key channels of agent integration, while the open-source project figma-console-mcp proves that the same access is technically achievable for free. This puts pressure on the monetization model for AI integration with design ecosystems: if the bypass becomes widespread, the company will face a choice — to move mandatory restrictions deeper, into the Plugin API and plugin licensing, or to reconsider the pricing of AI access. A wave of similar open-source bridges for other limited services is expected: the pattern is easily transferable. For builders, a free channel for agent-based product prototyping of "agent × design file" is opened: automated accessibility audits with auto-fixes directly in the file, design-to-code token pipelines according to DTCG, CSS, and Tailwind standards in CI, and edit digests for design review. The likely development over the coming years is one of two scenarios: either platforms close undocumented channels and monetize AI access as a separate service, or competition forces generous AI limits to become a standard part of the pricing plan. If the local plugin bridge survives, the fork of "cloud API with a paywall or local bridge" will become a standard part of design integrations, and the marginal cost of agent access to design data will drop to near zero — a class of agent-native tools on top of design files will grow.

Why this matters for users

If the monthly limit of the official Figma MCP has been exhausted, an agent can be connected to Figma for free, without upgrading the subscription. The procedure is as follows: the Figma desktop application is installed, the Desktop Bridge plugin is deployed via the figma-console-mcp npm package, after which the plugin is enabled in the Plugins → Development → Figma Desktop Bridge menu; macOS and Windows are supported. Then the agent works with the file directly: it edits frames, components, and layers, updates design tokens, takes screenshots for control, tracks edits, and gains access to FigJam and Slides — while the bridge itself does not keep count of calls. For those working from a browser, a cloud mode is available for web clients like Claude.ai, but with a reduced set of capabilities. This usually required a paid seat, but now the continuation of prototypes and internal pipelines after the free quota is exhausted is opened for free. A significant caveat: Code Connect is not included in the set, so for teams that link Figma with code, some scenarios will not be covered by this route.

What is still unknown / limitations

The claimed figures and promises require verification. "130 tools" is the project's self-description in release v1.46.0: this is the author's self-presentation, not independent verification, and the number of tools does not equal verified feature parity with the official server. The claim that "the Plugin API has no limits" is also stated by the project author and repository: there is no official position from Figma on this matter; this is a lack of a verifiable contract, and the route's behavior may change silently. For production-grade reliability, public data is lacking: nothing is known about latency, stability, quality of edits, the percentage of successfully completed agent runs, and how reliably the model selects the right tool from over a hundred options. Moreover, the architecture is tied to a running desktop application and the Plugin API, Code Connect is missing from the set, and in the event of a technical reaction from Figma to such bridges, the route may stop working. The realistic status today is pilots and non-critical pipelines, not a production replacement for the official server.

Sources

Author

Look at AI, editorial team