MCP server for Figma design creation via AI assistants
The Figma MCP server defines 5 shape/drawing creation tools with complete input schemas and reasonable descriptions. Naming is verb-first and clear (create_rectangle, create_text, etc.). However, descriptions are relatively generic (60-90 chars), lacking context on WHEN to use each tool or how they compose. Output schemas are completely undocumented, callers don't know what fields to expect from create_rectangle or create_text. Parameters lack detailed constraints (e.g., 'fills' is documented as 'Array of fill paints' but no guidance on structure). Error handling and recovery guidance are absent from the code. The server lacks any composition guidance, no indication of how to chain tools or what IDs are returned for downstream use. Security patterns (secret injection, audit trails) are not visible. Overall: functional definitions suitable for basic use, but lacking the LLM-optimization and downstream chaining support needed for production agentic workflows.
Creates an ellipse or circle in Figma. Use this for circular buttons, avatars, or decorative elements.
Creates a frame (container) in Figma. Use this to group elements or create screen areas.
Creates a line in Figma. Use this for dividers, connectors, or decorative elements.
Creates a rectangle in Figma. Use this for buttons, cards, backgrounds, or any rectangular shapes.
Creates a text element in Figma. Use this for headings, paragraphs, labels, or any textual content.
Output schemas completely undocumented. No documentation of what create_rectangle, create_text, etc. return. LLMs cannot plan downstream tool calls or extract required IDs (e.g., node ID for parent parameter in subsequent calls).
Parameter descriptions lack actionable constraints. 'Array of fill paints. Each paint can be a solid color, gradient, or image.' gives no structure, what fields does a paint object have? What type is 'color'? Should it be hex, RGB, or a Figma color ID?
Tool descriptions are generic and lack composition guidance. 'Creates a rectangle in Figma. Use this for buttons, cards, backgrounds, or any rectangular shapes.' does not explain when to use create_rectangle vs create_frame, or how to position/size shapes relative to each other.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 61 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
No error handling or recovery guidance in tool definitions. If x/y/width/height are invalid (e.g., negative values, out of canvas bounds), what error does the LLM receive, and what should it do next?
Complex nested structures (fontName, padding, cornerRadius) documented as object but with no inline schema or example. LLMs will struggle to construct valid payloads.
create_line has no required parameters (required: []), which may allow invalid calls (e.g., missing both endX and endY). Unclear if x/y/endX/endY are all optional or if some must be provided.
No documentation of what the 'parent' parameter ID format is or how to obtain it. If an agent needs to nest shapes, it must know the parent node ID, but create_frame, create_rectangle, etc. don't document what they return for this purpose.