SoftTech AI Registry
Figma · Figma Design

Figma agent design-file actions

The Figma agent's ability to perform persistent actions inside Figma Design through its agent interaction surface, covering layouts, component instances, styles, variables, content, images, layers, comments, design-system components, file-level state and multi-object selections, and the documented external MCP connector action surface the agent can invoke.

Recorded characteristics

Function
Figma documents an agent interaction surface inside Figma Design in which a user submits natural-language prompts and the agent performs work in the file. Figma's supported-feature documentation lists generation of screens and layouts, editing layout, editing component instances (properties, variants, overrides), applying styles, applying variables and switching modes, generating and editing content in bulk across multi-object selections, creating and editing images, converting images to vector layers, renaming layers, searching libraries, reviewing and acting on comments, providing design feedback, authoring design-system components, styles and variables including changes to main components and parent variants, performing file-level actions such as adding comments and modifying file metadata, and selecting objects on canvas. Figma documents some adjacent items as not supported or coming soon, including vector/Draw editing, prototyping and interactions, exporting assets and creating diagrams. Availability of individual items is subject to the plan and seat restrictions Figma documents and is not uniform across all plans and seats.
Data access
The agent operates on the Figma Design file the user is working in and can read file content, selections, libraries and comments in order to act on them. Where a user has configured an MCP connector, Figma documents that the agent can pull context from that external source into the design file. Figma does not establish in the reproduced documentation any broader data access beyond the file the user is in and the external sources the user has explicitly connected.
Actions
Can take actions
External actions
Yes
Human confirmation
Conditional
Permission basis
User permissions
Administrative control
Figma documents that administrators can enable or disable AI features at team, workspace or organization level, and that where AI is deactivated the agent is not available. Content training is documented as a setting separate from AI feature access. Figma also documents that administrators can disable connectors, and that on Organization and Enterprise plans administrators can publish custom MCP connectors for their organization.
Default state
Conditional
Availability
Figma documents the agent as being in beta, with rollout in waves beginning 20 May 2026, and states it is available on all plans. Full seats with edit access can ask the agent to make edits; users with View, Collab or Dev seats, or view-only file access, can chat with the agent without making edits, and can try the agent in Drafts. Figma documents that the agent is not available on Government or Education plans. Verified partner MCP connectors are documented as available on all plans; custom MCP connectors are documented as requiring a paid plan.
Licensing
Figma documents the agent as free to use during beta, with no AI credits consumed for prompts or actions during the beta period, and indicates that standard AI credit usage applies once the feature is generally available. Figma does not establish post-general-availability pricing in the reproduced documentation. Custom MCP connectors are documented as requiring a paid plan; the Starter plan is documented as lacking some listed features, but the detailed Starter entitlement boundary for individual agent features is not established.
External model or provider
Not established
Limitations and uncertainty
Human confirmation is recorded as conditional on the basis of the MCP connector path only. Figma documents that where a prompt requires a connector the agent shows which connector and tool it plans to call and asks for consent before running it, and that per-tool behaviour can be configured as Ask to run, Always run or Never run, so confirmation demonstrably varies by configuration. Figma does NOT establish, in the reproduced first-party documentation, whether native design-file edits have a separate pre-execution approval gate; native-edit confirmation behaviour is therefore not established. Submitting a prompt, authenticating a connector, holding edit permission, and the documented Undo control that reverts the most recent change are not treated as evidence of a separate human approval gate. On external actions: Figma documents that tools able to write to external sources are disabled by default and must be manually enabled, that connectors require configuration and per-person authentication through the external service, that external service permissions apply, and that connector and tool availability varies; the documented write path must not be generalised into a claim that every MCP connector or tool is writable. Figma sources differ in wording, with one describing the agent as beta and another as open beta; that difference is recorded rather than resolved. Also unresolved: post-general-availability AI credit and pricing behaviour; detailed Starter-plan entitlement boundaries for individual agent features; whether administrator-published custom connectors alter consent defaults; audit and logging granularity for agent-made changes; and any model or provider attribution, which Figma does not state for this capability.

Evidence