Figma tools
ChatGPTThe Figma app in ChatGPT generates beautifully designed presentations, and diagrams. Give the Figma app context about a presentation like a suggested outline or a goal, and it will generate a polished slide deck. The Figma app can also turn prompts into flow charts, Gantt charts, or other diagrams in FigJam. Open what you create in Figma to edit, share, and collaborate.
add_code_connect_map
ChatGPTMap a Figma node to a code component in your codebase using Code Connect. Use the nodeId parameter to specify a node id. Use the fileKey parameter to specify the file key. If a URL is provided, extract the node id and file key from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2 and the fileKey would be :fileKey. If the URL does not include node-id, ask the user for a node-specific URL. Do not pass an empty or guessed nodeId.
authorize_mcp_app
ChatGPTAuthorize the Figma app authentication session through the current OAuth connection. Called only by the app.
check_prompt_status
ChatGPTCheck one Figma Code generation started by send_prompt. Pass the fileKey and runId it returned. Read-only: does not start generation. Returns in_progress, needs_input, completed, failed, credits_exhausted, aborted, or unknown, with fileUrl and visible activity. While in_progress, wait at least pollAfterMs before checking again. Never resend a message to poll. For needs_input, mirror exactly one awaitingUserAnswers question in the client's native question UI, using question, options, multiSelect, and askUserQuestionContext verbatim. Call send_prompt again with continuation set to its resume fields and the answer as prompt, then poll its new runId. Only the initiating user with current file access can check a run. Status is retained for one hour after completion. Unknown means missing, expired, or unavailable outcome, not success: inspect the file in Figma and do not automatically retry generation.
create_generative_plugin
ChatGPTYou MUST load the figma-generative-plugins skill before calling this tool. If it is not installed, read skill://figma/figma-generative-plugins/SKILL.md with resources/read or get_figma_skill. Use this for requests to build, create, upload, or publish a “Figma plugin,” “generative plugin,” or “custom tool.” It creates a generative plugin in the account library; it does not install an existing Figma Community plugin. Creates a new generative plugin in the authenticated user's account library and returns its id. The plugin starts as a working scaffold — a runnable starter that draws a square — so it is a structural starting point, not a finished plugin: follow this call with the update tool to replace the scaffold's source with the behavior the user asked for. planKey names the plan that will own the plugin; take it from the plans list returned by whoami.
create_new_file
ChatGPTCreate a new blank Figma file. IMPORTANT: You MUST load the /figma-create-new-file skill BEFORE every call to this tool, if it exists. NEVER call this tool without loading that skill first if it exists. By default the file is placed in the authenticated user's drafts folder; If specified it can be placed inside a project. Use this tool when you need a new file to work with before calling use_figma. Returns the new file key and URL. Requires a planKey. If the user already provided a planKey, use it directly. Otherwise, call the whoami tool first to get the list of plans. If the user has one plan, use its "key" field. If multiple, ask the user which team or organization to use. Optionally accepts a projectId. If the URL is of the format https://figma.com/files/project/:projectId, https://figma.com/files/:orgId/project/:projectId, or https://figma.com/files/team/:teamId/project/:projectId then use the :projectId as the projectId.
create_shader
ChatGPTYou MUST load the figma-shaders skill before calling this tool. If it is not installed, read skill://figma/figma-shaders/SKILL.md with resources/read or get_figma_skill. Use this for requests to build, create, upload, or publish a “Figma shader,” “shader effect,” “shader fill,” “custom effect,” “custom fill,” or “procedural shader.” Creates a new shader effect or fill in the authenticated user's account library and returns its id. Set kind to effect for a shader that transforms the layer beneath it, or fill for a shader that generates its own pixels. The resource starts as a working scaffold, so follow this call with the update tool to replace the scaffold's source with the shader the user asked for. planKey names the plan that will own the shader; take it from the plans list returned by whoami.
download_assets
ChatGPTDownload assets from a Figma file for a single node: an exported render, the original source images, and SVGs of the vector layers. The response contains: (1) export — an exported image of the whole node; (2) rawImages — original uploaded source images (JPEG, PNG, GIF, WebP) found as fills anywhere in the node subtree (capped at 20); and (3) svgAssets — SVGs for the vector layers in the subtree that are best represented as SVG (icons, logos, simple illustrations), the same set get_design_context surfaces (capped at 20). Each raw image carries a format field with its actual image format (e.g. "png", "jpeg", "gif", "webp") so you can save it with the correct file extension; if the format cannot be determined it is reported as "original". Each svgAssets entry has format "svg". Call this tool for asset URLs get_design_context has not already provided: the export render of the whole node, a specific format or scale, or assets for a node you have not requested design context for. Export precedence: when you pass defaultFormat and/or defaultScale, those override the export settings configured on the node in Figma. When you omit them, the node-configured export settings are used if present, otherwise png at scale 1. Pass defaultFormat or defaultScale only when the user explicitly asks for a specific format, size, or resolution. For cross-file image transfer, use the raw image URLs with upload_assets. URLs are temporary — download promptly. Works on Figma design files (URL path /design/), Figma Slides (/slides/), and FigJam boards (/board/). Does NOT work on Figma Make files (/make/).
export_video
ChatGPTExport a Figma timeline node as an MP4 video. This tool only produces MP4 — GIF and animated SVG export are not supported yet. Renders the timeline server-side and returns a presigned download URL. The file stays available for ttlSeconds (defaults to 1 hour, clamped server-side to [30s, 7d]); use availableUntil in the response to know when it is deleted. Some renders finish in seconds, others take minutes; if the render hasn't finished within the handler budget, the response includes a jobId and status: "processing" — re-invoke export_video with { fileKey, jobId } after 10–15s to poll. Required: fileKey and either nodeId (to start a new export) or jobId (to poll). The nodeId must be the top-level frame that owns the timeline — in a design file a frame placed directly on the page or in a section, and in Slides the slide itself (not a layer inside it) — not a nested layer or sub-clip. If you animated a descendant, pass its containing top-level frame (the slide, in Slides); if get_motion_context returns a timelineCohorts entry, its rootNodeId is that frame. Use the quality field ("low"/"medium"/"high") to control output size.
generate_deck
ChatGPTGenerates polished and fully editable presentation decks in Figma Slides, suitable for a wide range of use cases including pitches, slideshows, portfolios, readouts, workshops, research summaries, moodboards, training materials, retrospectives, event recaps, and strategic reviews. This tool produces visually refined, ready-to-edit decks that can be customized for personal, creative, professional, corporate, and creative contexts. Use this tool when you need a slide deck that communicates ideas, findings, or proposals in a visually compelling way. Decks are optimized for storytelling, clarity, and design consistency—ideal for presenting to teams, clients, stakeholders, or audiences. Do not use this tool for designing application UIs, websites, flow diagrams, or standalone marketing assets. For FigJam diagrams, use the generate_diagram tool instead. For Buzz marketing assets, use generate_asset. This tool requires the following parameters to generate high-quality outputs: objectives, outline, style, color palette, use case, and theme. This tool does not retain chat history or conversational context beyond what you provide in the current request. To ensure best results, include all relevant details directly in your prompt, describing your goals and constraints clearly and completely. The prompt should be self-contained and include all relevant details with the context in mind but without referencing earlier prompts. This tool will return up to three unique slide deck options, each including a thumbnail and Figma Slides URL for editing within Figma. These generated decks serve as starting points for refinement, collaboration, and final presentation design.
generate_diagram
ChatGPTCreate a flowchart, decision tree, gantt chart, sequence diagram, state diagram, or entity relationship diagram in FigJam, using Mermaid.js. Generated diagrams should be simple, unless a user asks for details. This tool also does not support generating Figma designs, class diagrams, timelines, venn diagrams, or other Mermaid.js diagram types. This tool also does not support font changes, or moving individual shapes around -- if a user asks for those changes to an existing diagram, encourage them to open the diagram in Figma. If the tool is unable to complete the user's task, reference the error that is passed back. Do not use the create_new_file tool prior to creating a diagram using this tool; generate_diagram creates its own files.
generate_figma_design
ChatGPTCapture, import, or convert a web page (by URL) or HTML into an existing Figma design file. Use this tool when the user wants to send, import, capture, screenshot, or push a webpage into Figma. REQUIRES an existing fileKey — if the user does not already have a Figma file, first call create_new_file (load the figma-create-new-file skill for the plan-resolution contract) and reuse the returned file_key here. Works with both local dev servers (localhost) and external websites. For LOCAL projects: explore the user's codebase to identify the page, dev server command, and URL before calling. Call this tool with a fileKey (no captureId) to get the capture script + a captureId, then poll with the captureId every 5 seconds (up to 10 times) until status is 'completed'. Each capture ID is single-use. USING generate_figma_design WITH use_figma: - For web apps, use BOTH tools in parallel: run this tool (generate_figma_design) to capture a pixel-perfect layout, and use use_figma with search_design_system to build the screen from design system components. - Once both complete, refine the use_figma output to match the pixel-perfect layout from this tool's capture, then delete this tool's output (it was used as a layout reference only). - For non-web apps (e.g. iOS), use use_figma only. - For updating/syncing a page already in Figma, use use_figma only.
generate_image
ChatGPTGenerate a raster image from a text prompt using state-of-the-art image models. Width and height guide the aspect ratio; the model chooses a supported output size, so the image pixel dimensions may differ from the requested values. The tool returns a URL serving the raw image bytes of a PNG or JPEG image. IMPORTANT: The URL expires after 7 days. The user MUST download the image before the URL expires to keep a copy. This is a paid feature — each call consumes Figma AI credits against the user's plan (different models have different per-call costs).
get_chat_threads
ChatGPTList the Figma Code conversation threads you can send a prompt to, so you can choose one before calling send_prompt. Read-only: does not start generation. A standalone Code file returns that file's threads. A Design file returns the file agent's threads, or, when nodeId names a code layer, that layer's own threads instead. A nodeId that is not a code layer does not narrow the lookup: the file agent's threads are returned and scope comes back as "file_agent" rather than "code_layer", so check scope before assuming a result belongs to the node you asked about. Returns threads newest first with threadId, type, updatedAt, and title when the conversation has one — prefer the title over recency when choosing between several. An empty list means no conversation exists yet; send_prompt starts one. Only threads you can see are listed: a file-scoped thread is visible to anyone with file access, a private thread only to its owner.
get_code_connect_map
ChatGPTGet a mapping of {[nodeId]: {codeConnectSrc: e.g. location of component in codebase, codeConnectName: e.g. name of component in codebase} E.g. {'1:2': { codeConnectSrc: 'https://github.com/foo/components/Button.tsx', codeConnectName: 'Button' } }. Use the nodeId parameter to specify a node id. Use the fileKey parameter to specify the file key. If a URL is provided, extract the node id and file key from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2 and the fileKey would be :fileKey. If the URL does not include node-id, ask the user for a node-specific URL. Do not pass an empty or guessed nodeId.
get_code_connect_suggestions
ChatGPTGet AI-suggested strategy for linking a Figma node to code components via Code Connect. Workflow: call this tool → review suggestions with the user → call send_code_connect_mappings to save the approved mappings. Use the nodeId parameter to specify a node id. Use the fileKey parameter to specify the file key. If a URL is provided, extract the node id and file key from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2 and the fileKey would be :fileKey. If the URL does not include node-id, ask the user for a node-specific URL. Do not pass an empty or guessed nodeId.
get_context_for_code_connect
ChatGPTGet structured component metadata including properties, variants, and descendant tree for a Figma component or component set. Returns property definitions with types and variant options, and a tree of descendant instances and text nodes with their property references. Designed for creating Code Connect template files. Use the nodeId parameter to specify a node id. Use the fileKey parameter to specify the file key. If a URL is provided, extract the node id and file key from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2 and the fileKey would be :fileKey. If the URL does not include node-id, ask the user for a node-specific URL. Do not pass an empty or guessed nodeId.
get_design_context
ChatGPTGet design context for a Figma node — the primary tool for design-to-code workflows. Returns reference code, a screenshot, and contextual metadata that should be adapted to the target project. See the server instructions for how to interpret and adapt the response. Use the nodeId parameter to specify a node id. Use the fileKey parameter to specify the file key. If a URL is provided, extract the node id and file key from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2 and the fileKey would be :fileKey. If the URL does not include node-id, ask the user for a node-specific URL. Do not pass an empty or guessed nodeId. If the URL is of the format https://figma.com/design/:fileKey/branch/:branchKey/:fileName then use the branchKey as the fileKey. If the URL is of the format https://figma.com/make/:makeFileKey/:makeFileName then use the makeFileKey to identify the Figma Make file. Only for Figma Make files (URLs containing /make/), and only when calling get_design_context, assume the nodeId is 0:1. The response will contain a code string and a JSON of download URLs for the assets referenced in the code. It will also include a screenshot of the node for context by default.
get_figjam
ChatGPTGenerate UI code for a given FigJam node in Figma. Use the nodeId parameter to specify a node id. If no node id is provided, use 0:1 which is the root node ID. Use the fileKey parameter to specify the file key. If a URL is provided, extract the node id from the URL, for example, if given the URL https://figma.com/board/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2 and the fileKey would be :fileKey. IMPORTANT: This tool only works for FigJam files (URL path /board/), not other Figma files.
get_generative_plugin
ChatGPTReads a generative plugin from the account library by id (from list_generative_plugins), returning its name, description, owner, version, and a manifest of its source files as { filename, bytes, uri }. Owner is the authenticated user's email when they own the plugin, or a public publisher handle otherwise. Read each file's contents from its uri as an MCP resource (contents are not inlined here). Only set includeSource to true to add source to each file when the MCP client cannot read MCP resources. Pass an optional version (commit SHA) to read a specific build; defaults to the latest.
get_libraries
ChatGPTGet the design libraries associated with a Figma file. Returns two lists: (1) libraries currently added to the file (subscribed), and (2) libraries available to add (community UI kits and organization libraries). Each library includes its name, library key, description, and source type. The organization libraries portion of libraries_available_to_add is paginated — when the response includes a libraries_available_to_add_next_offset value, pass it back via the offset parameter to fetch the next page. Use the library keys from the response to scope searches with search_design_system by passing them as includeLibraryKeys.
get_metadata
ChatGPTIMPORTANT: Always prefer to use get_design_context tool. Get metadata for a node or page in the Figma desktop app in XML format. Useful only for getting an overview of the structure, it only includes node IDs, layer types, names, positions and sizes. You can call get_design_context on the node IDs contained in this response. Use the nodeId parameter to specify a node id, it can also be the page id (e.g. 0:1). IMPORTANT: This tool only works for Figma design files (URL path /design/). It is NOT supported for FigJam (/board/) or Slides (/slides/) files. This tool is not supported for Figma Make Files (URLs containing /make/). The nodeId parameter is optional: when omitted, the tool returns a list of the top-level pages (guid + name) in the document instead of an XML dump — use this when you don't yet know which page or node to drill into. If the URL includes node-id, extract it and pass it as nodeId; for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2. If the URL does not include node-id, do not set nodeId; omit the field so the tool lists top-level pages. Do not pass an empty or guessed nodeId. If the URL is of the format https://figma.com/design/:fileKey/branch/:branchKey/:fileName then use the branchKey as the fileKey.
get_motion_context
ChatGPTGet keyframe animation data for a Figma node. Returns animated-node inventory, keyframe tracks with easing curves, pre-computed CSS/@keyframes and motion.dev code snippets, and timeline coordination hints for recursive calls. Use after get_design_context for motion-aware code generation. Use the nodeId parameter to specify a node id. Use the fileKey parameter to specify the file key. If a URL is provided, extract the node id and file key from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2 and the fileKey would be :fileKey.
get_screenshot
ChatGPTGenerate a screenshot for a given node or the currently selected node in the Figma desktop app. Works on Figma design files (URL path /design/), FigJam boards (/board/), and Figma Slides (/slides/). The optional maxDimension parameter (positive integer, max 65536, default 1024) caps the longer edge of the rendered PNG in pixels — increase it when you need to inspect fine detail, decrease it for thumbnails or to save context. The JSON metadata entry in the response includes both width/height (the rendered PNG size) and original_width/original_height (the node's natural canvas size before any clamping), so callers can decide whether to re-request at a higher maxDimension. Use the nodeId parameter to specify a node id. nodeId parameter is REQUIRED. Use the fileKey parameter to specify the file key. fileKey parameter is REQUIRED. If a URL is provided, extract the file key and node id from the URL. For example, if given the URL https://figma.com/design/pqrs/ExampleFile?node-id=1-2 the extracted fileKey would be pqrs and the extracted nodeId would be 1:2. If the URL does not include node-id, ask the user for a node-specific URL. Do not pass an empty or guessed nodeId. This tool is not supported for Figma Make Files (URLs containing /make/). By default this tool returns a short-lived URL to the screenshot plus curl instructions for downloading the PNG — the URL+curl path is strongly preferred because it uses far fewer tokens than embedding the image inline. The enableBase64Response parameter defaults to false. Only set enableBase64Response: true when the agent cannot fetch URLs (no shell access, no HTTP client, or a sandboxed environment that blocks outbound requests); when set, an inline base64 image entry is appended to the response in addition to the URL and curl instructions. If the URL is of the format https://figma.com/design/:fileKey/branch/:branchKey/:fileName then use the branchKey as the fileKey.
get_shader
ChatGPTReads a shader effect or shader fill from the account library by id (from list_shaders), returning its name, description, owner, type, version, and a manifest of its source files as { filename, bytes, uri }. Owner is the authenticated user's email for their shaders, or figma for first-party shaders. Read each file's contents from its uri as an MCP resource. Only set includeSource to true to add source to each file when the MCP client cannot read MCP resources. Pass an optional version (commit SHA) to read a specific build; defaults to the latest.
get_variable_defs
ChatGPTGet variable definitions for a given node id. E.g. {'icon/default/secondary': #949494}Variables are reusable values that can be applied to all kinds of design properties, such as fonts, colors, sizes and spacings. Use the nodeId parameter to specify a node id. Extract the node id from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2. If the URL does not include node-id, ask the user for a node-specific URL. Do not pass an empty or guessed nodeId. This remote tool requires a concrete node target. This tool is not supported for Figma Make Files (URLs containing /make/). If the URL is of the format https://figma.com/design/:fileKey/branch/:branchKey/:fileName then use the branchKey as the fileKey.
list_file_components_for_code_connect
ChatGPTList every component and component set PUBLISHED to a Figma file's library, with the cross-component dependency graph needed to plan Code Connect in bulk. Only published components are returned (unpublished/local-only components are omitted). Returns one entry per component with its properties (exhaustive variant options, defaults, SLOT descriptions), page and asset/library membership, instance count, and direct dependencies (each flagged internal vs. external). Child instances and text layers include selector paths, property references, visibility, and owning component-set variants. Instance records include resolved master identity and property definitions; text records include sample content from the placed layer. Equivalent records across variants are combined with one sample nodeId. Instance-swap children are included but are not dependency edges. Unlike get_context_for_code_connect — which returns the deep descendant tree for one known component — this returns the flat whole-file graph for dependency-ordered, batched template generation. Takes only a file key (no node id). Use the fileKey parameter to specify the file key. If a URL is provided, extract the file key from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName, the extracted fileKey would be :fileKey. If the URL is of the format https://figma.com/design/:fileKey/branch/:branchKey/:fileName then use the branchKey as the fileKey.
list_file_shaders
ChatGPTLists the shader effects and shader fills used in a Figma file. Returns each shader as { id, name, description, type, version, published, truncated, files }, where type is "effect" (post-effect that samples an input raster) or "fill" (generates pixels directly), published indicates whether the shader is a published library version, and files is a manifest of its authored source files as { filename, uri }. When truncated is true, the manifest hit its 10,000-file safety cap and is not exhaustive. The top-level truncated field is true when a referenced shader could not be returned, including when the file references more than the 100 returned shaders. Read each file's contents from its uri as an MCP resource (contents are not inlined here). Requires view access to the file. Use this to inspect shaders in a file you can open, including ones you do not own. Source reads are pinned to the exact shader version referenced by the file.
list_generative_plugins
ChatGPTLists the generative plugins in the authenticated user's account library, including Figma's first-party plugins. Returns each plugin's id, name, description, and owner (plus a nextCursor when more pages exist). Owner is the authenticated user's email for their plugins, or a public publisher handle otherwise. Use the id with get_generative_plugin to read a plugin's source. Generative plugins are runnable tools that modify the canvas, distinct from shader effects and shader fills.
list_shaders
ChatGPTLists the shader effects and shader fills in the authenticated user's account library. Returns each shader's id, name, description, owner, and type (effect or fill), plus a nextCursor when more pages exist. Owner is the authenticated user's email for their shaders, or figma for first-party shaders. Use the id with get_shader to read either shader type's source.
open_figma_mcp_app_in_thread
ChatGPTOpen Figma in the host application thread view.
publish_to_gallery
ChatGPTPublish a completed standalone generic Figma Code file as a site and submit it to Figma Gallery. Supports eligible standalone files, not Design frames or code layers in Design files. Requires a completed send_prompt run for the same file, successful production-deploy verification after final code changes, and user authorization to publish to Gallery. An explicit Gallery publishing request is sufficient; otherwise obtain human confirmation. Authorization can precede setup and survives setup repairs. Honor declines and revocation; mentioning Gallery in a question or negation is not authorization. Importing source alone does not satisfy these prerequisites. Do not publish if completion or verification is unconfirmed. Load share-to-figma-gallery before publishing and follow its workflow. Pass the accepted title and description as plain text, preserving user edits. Pass the same fileKey and completed runId from send_prompt, checked with check_prompt_status. Never supply repository URLs, refs, commits, or Make node IDs. This tool waits for Gallery publishing. On success, report success and include the Gallery post URL when galleryPostUrl is returned; otherwise report success without a URL. Do not expose file, site bundle, or Gallery publish status IDs. Do not automatically retry failed or unconfirmed publication.
report_mcp_app_failure
ChatGPTReport a bounded Figma app failure category and optional numeric error code. Called only by the app.
search_design_system
ChatGPTSearch for design system assets (components, variables, and styles) based on a text query. Returns matching assets from all design libraries. Use this when you need to find specific components, variables (e.g. colors, spacing tokens), or styles from design libraries.
send_code_connect_mappings
ChatGPTSave multiple Code Connect mappings in bulk. Use after get_code_connect_suggestions to confirm and save approved mappings. Use the nodeId parameter to specify a node id. Use the fileKey parameter to specify the file key. If a URL is provided, extract the node id and file key from the URL, for example, if given the URL https://figma.com/design/:fileKey/:fileName?node-id=1-2, the extracted nodeId would be 1:2 and the fileKey would be :fileKey. If the URL does not include node-id, ask the user for a node-specific URL. Do not pass an empty or guessed nodeId.
send_code_to_figma
ChatGPTImport a local project into Figma Code. Requires edit access to a standalone Code or Design file. Does not create native Design shapes. Before choosing or creating a destination file, load figma-code-workflows. Target: standalone Code uses its empty main layer, including a starter. Design creates a new layer on pageId (default: first page). For populated Make, create a blank file. Same-target repair requires returned repairFrom: keep fileKey, seedId and makeNodeId; upload the corrected complete project. Changed or successful imports cannot be replaced. Legacy layers are unsupported. Start with the destination, omitting upload, to request instructions. Follow the returned source-configuration and archive guidance; local dependency installation, dev servers, HTTP checks and builds are not required. Static HTML/CSS/JS needs no setup scripts. Create the archive at nextStep.localArchivePath, then run nextStep.uploadCommand. Only after upload succeeds, call send_code_to_figma with nextStep.thenCall.arguments unchanged within one hour. upload.archivePath is a Figma storage path, never a local path. Keep makeNodeId/createLayer and any pageId/repairFrom unchanged. Do not resume failed uploads or retry ambiguous failures. Returns importReceipt, fileUrl, fileKey and nodeId. needs_setup preserves valid source with setupError; follow returned recovery instructions. Unsafe archives/access failures still fail. For import_pending, follow the continuation. Imported does not verify installation, preview or publishing. Open the file to start preview. For edits, use send_prompt with fileKey; also pass nodeId for a Design code layer. Use it for repair only when the user explicitly asks Figma to repair the app. check_prompt_status is for generation.
send_prompt
ChatGPTSend a prompt to Figma Code to create or edit a code-backed app. Code agent only; no Design Agent or native shapes. Requires edit access. Standalone Code files use their main layer. In Design, supply nodeId to edit a code layer; omit nodeId and threadId to create one on pageId (default: first page). Omit threadId to reuse the default conversation in Design or start a fresh one in standalone Code. Reuse the returned threadId for follow-ups, with its nodeId in Design. Legacy layers are unsupported. To attach local files, set attachment_count to 1–10 with your prompt. Returns one upload slot per file without starting generation. Upload all files (up to 5 MiB each, within one hour), then follow nextStep.thenCall, filling in filenames and MIME types in slot order. Keep its prompt and target fields unchanged. The resumed call attaches the files to the conversation and runs the Code agent. Works on populated targets; does not replace the repository or scaffold the uploads. Omit attachment_count or set 0 for a prompt without uploads. Failed uploads must not be resumed. For project initialization and scaffolding, use send_code_to_figma when available; follow up with send_prompt if needed. After generation is accepted, returns fileUrl, nodeId, fileKey, threadId and runId. Wait pollAfterMs, then poll check_prompt_status with fileKey and runId. Send one message per target at a time. Do not automatically retry ambiguous failures. For awaitingUserAnswers, mirror one question in the client's native question UI using question, options, multiSelect and askUserQuestionContext verbatim. Call send_prompt with continuation set to its resume fields and the answer as prompt. Omit nodeId, pageId, threadId and attachment fields.
update_generative_plugin
ChatGPTYou MUST load the figma-generative-plugins skill before calling this tool. If it is not installed, read skill://figma/figma-generative-plugins/SKILL.md with resources/read or get_figma_skill. Use this when the user asks to update, revise, or republish an existing Figma plugin, generative plugin, or custom tool in their account library. Updates an existing generative plugin in the authenticated user's account library. Provide the plugin id, existing authored files to replace, optional name and description metadata, and a required Git commit message describing the change. The update is built, versioned, and deployed; record the new version when the response includes it. Use create_generative_plugin first when no plugin exists.
update_shader
ChatGPTYou MUST load the figma-shaders skill before calling this tool. If it is not installed, read skill://figma/figma-shaders/SKILL.md with resources/read or get_figma_skill. Use this when the user asks to update, revise, or republish an existing Figma shader, shader effect, shader fill, custom effect, custom fill, or procedural shader. Updates an existing shader effect or fill in the authenticated user's account library. Provide its id, matching kind, existing authored files to replace, optional name, description, animation, and mouse metadata, and a required Git commit message describing the change. The update is built, versioned, and deployed; the response includes the new version. Use create_shader first when no shader exists.
upload_assets
ChatGPTUpload assets (images, etc.) into a Figma file. Call with a "count" to get that many single-use upload URLs. POST raw asset bytes to each URL with the correct Content-Type header (e.g. image/png, image/jpeg). Set batchCommit to true so all assets can commit and place in one file operation. If a commitUrl is returned, after all uploads succeed, call it exactly once to commit and place all assets. If no commitUrl is returned, each upload URL commits and places automatically. For a single image with nodeId, sets it as a fill on that existing node. Without nodeId, creates new frames with image fills on the current page. Supports PNG, JPG, GIF, and WebP formats. Max 10MB per asset. Works on Figma design files (URL path /design/), FigJam boards (/board/), and Figma Slides (/slides/).
use_figma
ChatGPTCreate, edit, generate, or sync any design in Figma — UIs, screens, mockups, components, frames, variables, styles, text, images, layouts, and design systems. This is the general-purpose tool for writing to Figma; it works by running JavaScript via the Figma Plugin API. Works on Figma design files (URL path /design/), FigJam boards (/board/), and Figma Slides (/slides/). IMPORTANT: Before calling this tool, load figma-use guidance. Prefer the /figma-use skill if available; otherwise read the skill://figma/figma-use/SKILL.md MCP resource. Skipping this causes common, hard-to-debug failures. Use this tool when the user wants to: - Create or generate a design, screen, UI, or mockup in Figma — from scratch, from intent, or from code - Update, edit, or sync an existing Figma design - Generate or sync Figma designs from source code - Set up or modify design tokens, variables, or styles - Build or extend a design system or component/variant library - Fix layout, spacing, auto-layout, or fill/hug issues - Add component descriptions or Code Connect metadata to nodes - Review or fix accessibility, contrast, typography, or visual polish - Inspect or query node properties programmatically CHOOSING BETWEEN use_figma AND generate_figma_design: - Default to this tool (use_figma) for all Figma write operations. - generate_figma_design is the exception: use it ONLY when capturing a web app page or view into Figma for the first time. For web apps, run both tools in parallel — generate_figma_design captures a pixel-perfect screenshot, use_figma builds the screen from imported design system components, then refine use_figma against the screenshot. - For non-web targets (iOS, Android, generic UI) and from-scratch designs, use this tool only. - For updating or syncing a Figma page already captured into Figma, use this tool — even if the source code has changed. GOTCHAS: - For the font "Inter", the style is "Semi Bold" (with a space), not "SemiBold". Same for "Extra Bold" not "ExtraBold". - Setting figma.currentPage is not supported. Use await figma.setCurrentPageAsync(page) instead. - getPluginData / setPluginData / getPluginDataKeys are NOT supported here (web-only, require a plugin manifest id). Use getSharedPluginData(namespace, key) / setSharedPluginData(...) / getSharedPluginDataKeys(namespace) instead. Pick a stable namespace (>=3 chars, alphanumeric/_/.) unique to your integration. REUSE THE DESIGN SYSTEM FIRST: Before creating components, styles, or tokens from scratch, call search_design_system to find existing matches. Import component matches via importComponentByKeyAsync or importComponentSetByKeyAsync rather than recreating them. Reuse existing variables and styles rather than defining new ones.
weave_cancel_tool_run
ChatGPTCancels one or more in-progress runs of a Weave tool (a published Weave workflow). Pass the recipeId of the tool and, optionally, the runIds to cancel (from weave_run_tool); omit runIds to cancel all of the user's currently running runs for that tool. Use this to stop a run the user no longer wants. Cancellation cannot be undone.
weave_find_model
ChatGPTFinds a Weave AI model by name so it can be run directly with weave_run_model — no Weave tool needed. Use this when the user names a model ("run nano banana 2 on this image", "make a video with veo 3"). A model is not a Weave tool: for a published Weave workflow use weave_list_tools and weave_run_tool instead. Returns the model with an approximate cost and the contract to run it with — its inputs and tunable params, each with a type, whether it is required, and any allowed values — or a short list of candidates to choose between when the name matches more than one model.
weave_get_model_run_output
ChatGPTGets the output and status of Weave model runs started with weave_run_model. Pass the predictionIds those calls returned. This is for model runs only — for a run of a Weave tool (a published Weave workflow) use weave_get_tool_run_output instead. Poll this after weave_run_model to track progress and read the output. The response also includes curl instructions for downloading the outputs — the URL+curl path is strongly preferred because it uses far fewer tokens than embedding media inline.
weave_get_tool_inputs
ChatGPTGets the input contract of a Weave tool (a published Weave workflow) — the inputs you fill in to run it. Pass the recipeId (from weave_list_tools, or the <id> in a pasted Weave URL like app.weavy.ai/tool/<id> or app.weavy.ai/flow/<id> — that <id> is the recipeId). A pasted Weave URL is enough to inspect and run the tool right here — do not open a browser or use browser automation for Weave. Call this before weave_run_tool to learn what to send. Returns the tool version (pass it back to weave_run_tool), an outputs summary (what the tool produces), and a flat inputs list. Each input has a nodeId (the key you send in weave_run_tool), name, type (text, integer, boolean, select, seed, image, video, audio, 3D, color, or any), default (its current value), and required (true only when there is no current value, so you must provide one), plus options (for select), range, isIterator, and description.
weave_get_tool_run_output
ChatGPTGets the output and status of runs of a Weave tool (a published Weave workflow). Pass the recipeId of the tool and the runIds returned by weave_run_tool; omit runIds to get the tool's most recent run. Returns each run's status (RUNNING, COMPLETED, FAILED, or CANCELED), progress, any error, and — when complete — a link to every output the run produced. Poll this after running a tool to track progress and read its output. Each output is a JSON entry with a url, its type (e.g. image, video), and, when known, width/height/format. The response also includes curl instructions for downloading the outputs — the URL+curl path is strongly preferred because it uses far fewer tokens than embedding media inline.
weave_list_tools
ChatGPTLists the Weave tools the authenticated user can run — published Weave workflows — in their active Weave workspace: their own, those shared with the workspace, and those shared with them directly. Here "tool" means a Weave tool (a published Weave workflow), not an agent/MCP tool. Use this when the user wants to see, browse, or choose from the Weave tools available to them. Returns the most recently updated tools and the total number available, each with its name, who created it, when it was last updated, and a link to open it in Weave. If the user already pasted a specific Weave URL (app.weavy.ai/tool/<id> or app.weavy.ai/flow/<id>), you already have its recipeId — that <id> — so skip this and go straight to weave_get_tool_inputs/weave_run_tool instead of opening a browser.
weave_run_model
ChatGPTRuns a Weave AI model directly — no Weave tool needed — and returns a prediction id; poll it with weave_get_model_run_output. Call weave_find_model first for the model id and its contract. Running spends the user's Weave credits, so it is gated: a call without acknowledgedCost only quotes and spends nothing, returning inputs_required or cost_confirmation_required with instructions to follow. Always show the user the cost and get an explicit Approve/Cancel (a structured prompt, not free text) before echoing acknowledgedCost to run — every run, including reruns. Ask the user for any input you don't have; never invent one.
weave_run_tool
ChatGPTRuns a Weave tool (a published Weave workflow) and returns run ids; poll them with weave_get_tool_run_output. A pasted Weave URL (app.weavy.ai/tool/<id> or app.weavy.ai/flow/<id>) is enough — never open a browser or use browser automation for Weave. Call weave_get_tool_inputs first to learn the inputs. Running spends the user's Weave credits, so it is gated: a call without acknowledgedCost only quotes and spends nothing, returning status: "inputs_required" or cost_confirmation_required with instructions to follow. Always show the user the cost and get an explicit Approve/Cancel (a structured prompt, not free text) before echoing acknowledgedCost to run — every run, including reruns — and confirm any auto-filled inputs (values you did not set that the run will use). Include the response's costDisclosure string as its own sentence after the question, not folded into it, the first time you quote a cost in a conversation; omit it after that. For an image/video input, pass a reachable https URL directly; only upload a local file with weave_upload_asset and pass the asset object it returns. If the user wants an input you did not get from weave_get_tool_inputs, do not omit it or fold it into another input like the prompt — re-fetch weave_get_tool_inputs (its inputs can change) and set it.
weave_upload_asset
ChatGPTUploads a local image or video file to Weave and returns the asset object to pass as the value for an image/video input in weave_run_tool. Returns a submitUrl and a token: POST the file to the submitUrl as multipart/form-data with a file field and the token in an X-Weave-Upload-Token header (e.g. curl -F "file=@/path/to/image.png" -H "X-Weave-Upload-Token: <token>" "<submitUrl>"); that POST returns the asset object. If you already have a reachable https URL, do not upload it — pass it directly to weave_run_tool. If the media exists only in this conversation (you have no path or URL for it), ask the user to save it and give you a path or URL. The token expires in about 10 minutes.
whoami
ChatGPTReturns information about the authenticated user. If you are experiencing permission issues with other tools, you can use this tool to get information about who is authenticated and validate the right user is logged in.
add_code_connect_map
Claude
