Skip to main content
Glama
Figma Write Bridge MCP

Figma Write Bridge MCP

by kicholiz

Figma Write Bridge

Figma Write Bridge lets a local tool (like an AI assistant or a script) safely “write” into your open Figma file by connecting a Figma plugin to a local bridge server on your computer.

This repo contains:

  • figma-plugin/ — the Figma plugin (connects your Figma file to the bridge)

  • mcp-server.js — the local bridge server (runs on your computer)

  • package.json — Node.js setup (starts the server)

  • skills/figma-write-bridge/ — agent skill; zip this folder and upload the zip to your AI agent (see below)


What You’ll Use It For

  • Let an assistant generate frames, text, shapes, styles, and structured layouts in your Figma file.

  • Keep control: nothing happens unless your Figma file is open and the plugin is connected.

Related MCP server: Figma MCP Server

What It Can Do

  • Frames, text, shapes, sections, and vectors — create and edit nodes, including sections (create_section, set_section_properties to flip SECTION/VIEWPORT), vectors from SVG paths, and boolean groups.

  • Move anything anywhere — move_node (absolute x/y or relative dx/dy on freeform parents; inside auto layout this is rejected unless you pass ignoreAutoLayout: true for an overlay), reparent_node / insert_child (move any node into/out of frames, sections, groups, auto-layouts, and slots, with an index for order — omit index to append), append_to_slot (into component slots; children join the slot’s layout flow), move_node_to_page (cut or copy to another page), and clone_node_into_parent (copy into any container).

  • Resize to fit — resize_to_fit scales any layer to fit inside a target layer (fit: "contain" letterboxes, fit: "cover" fills/crops, both aspect-preserving and centered), or shrink-wraps a container tightly to its own children when no targetNodeId is given.

  • Find nodes by query — find_nodes filters the whole document server-side by type, name/text, fill color, style/variable binding, instance overrides, and more (all combinable), so a search like "red button instances" or "hardcoded colors with no style" returns only the matches instead of a full tree dump.

  • Layout & structure — auto layout, padding/spacing/alignment, layout grids, one-call grid generators (generate_grid), and layout helpers (distribute_nodes, arrange_children). Those helpers skip auto-layout children instead of pulling them out of the flow — use set_item_spacing / set_axis_align / insert_child on stacks. Primary-axis alignment includes SPACE_AROUND and SPACE_EVENLY. create_frame defaults to no fill; nested auto-layout wrappers omit width/height so they do not ship at 320×200.

  • Grid auto-layout — set_grid_layout turns a frame into a true two-dimensional GRID container (row/column counts, gaps, and per-track FLEX/FIXED/HUG sizing), set_grid_child_position places each child by cell, span, and in-cell alignment, get_grid_layout reads the whole arrangement back, and reorder_grid_tracks moves entire rows or columns with their children.

  • Fills & effects — solid/gradient/image fills (from a URL, base64, or a local file path via localPath), and the full effect set: drop/inner shadows, normal and progressive blurs, plus NOISE, TEXTURE, and GLASS. Apply existing styles or variable-bound colors. create_frame defaults to no fill (transparent layout containers); pass fillHex only for visible surfaces. set_fill_color / set_stroke_color support clear: true.

  • Strokes — set_stroke_color sets color/weight plus strokeAlign, dashPattern, strokeCap, strokeJoin, or applies a paint styleId; prefer apply_stroke_style / bind_color_variable_to_stroke when the file has tokens.

  • Text styling — apply existing text styles or set font/font-size/line-height/letter-spacing/case/alignment directly, with variable binding, paragraph-level textWrapStyle (BALANCE/PRETTY), textTruncation + maxLines, and variable-font variationSettings (after get_font_variation_axes). Generate a whole type scale from a base size + ratio with create_typography_scale.

  • Design tokens — export local variables as a W3C-style Design Tokens JSON (export_tokens) and import a tokens JSON into variables + paint styles (import_tokens).

  • Style guides & palettes — extract a usage style guide (get_style_guide: colors, fonts, sizes, spacing), list fonts used (get_font_list), and generate tonal color palettes with swatches/styles/variables (generate_palette).

  • Components — create/import components and instances (imports accept a name to rename the main node), batch-convert frames into a variant component set (extract_component_set), and move/copy a local component to another open file's channel with move_component_to_file. get_local_components is paged (limit/offset + total).

  • Undo/redo — snapshot-based undo / redo for the most recent mutating actions, shared between the agent's tools and Undo/Redo buttons in the plugin UI itself (best-effort; cannot restore deleted nodes or structural changes).

  • Pages — create, rename, duplicate (auto-names like Name 2 or takes a name), reorder, switch, and delete pages; create_page / duplicate_page accept activate: true to switch to the new page.

  • Bulk & template work — bulk_rename, bulk_update, replace_all_instances, and page duplication.

  • Variables & themes — variable types cover COLOR, FLOAT, STRING, BOOLEAN, and the motion types EASING and TIMING, so animation curves and durations can be tokenized alongside color and spacing. Read variable values in every mode (list_variables with includeValues: true for the catalog, get_variable for one variable in detail with alias-resolved values per mode), create/rename/delete variable modes and collections, write values into any mode (set_variable_values(valuesByMode)), and theme-switch whole frames/pages with set_variable_mode. Catalog listings are paged to keep responses token-cheap: list_variables / get_local_components / get_styles take limit/offset (default 500) and return total so you can page through large catalogs.

  • Prototyping — frame-to-frame links with typed transitions and easing (set_transition_reaction, set_smart_animate_reaction), multi-action triggers (set_reactions, upsert_reaction), overlays, flows, and start points.

  • Motion (timeline animation) — animate properties over a frame's timeline rather than between frames: manual keyframe tracks for transform, opacity, radius, size, spacing and path trim, plus indexed fill/stroke/effect color tracks (set_keyframe_track), reusable animation styles (list_animation_styles / apply_animation_style), and timeline length (set_timeline_duration). Read it all back with get_motion. Requires Figma to have enabled Motion for your account; the tools say so plainly when it hasn't.

  • Shaders — list_shaders enumerates shader effects/fills available to the file. import_shader_by_id materializes one into the file; apply_shader applies it as an effect, fill, or stroke (with properties keyed by name or definition id).

  • Variable fonts — get_font_variation_axes reads OpenType axes; create_text / set_text_style / create_text_style accept variationSettings (e.g. { wght: 550 }).

  • Live push events — subscribe to selectionchange / documentchange so the agent can react to your selection or canvas without polling.

  • Channel dashboard — list_channels shows which file each connected channel belongs to.

  • REST API extras — file JSON, image downloads, bulk frame exports, file comments, and component search (with FIGMA_TOKEN). Video export of Motion frames (MP4/GIF/WEBM via export_node_as_image) is plugin-side and does not need a token.


Prerequisites

  • Figma Desktop app (recommended for local plugin + localhost connections)

  • Node.js (LTS) installed on your computer


Install the Figma Plugin

  1. Open Figma Desktop.

  2. Go to Plugins → Development → Import plugin from manifest…

  3. Select this file:

    • figma-write-bridge\figma-plugin\manifest.json

  4. The plugin appears as “Figma Write Bridge (Local)” under Plugins → Development.


Start the Local Bridge Server

In a terminal (PowerShell is fine), run the following, replacing the path with the actual path to this repo on your computer:

cd "C:\path\to\figma-write-bridge"
npm install
npm start

By default, the server listens on:

  • ws://localhost:8787

Keep this terminal window open while you use the bridge.


Add to Your AI Agent (MCP config)

If your AI agent supports MCP tool servers, you can register this bridge so the agent can call Figma tools.

  1. Make sure the Figma plugin is connected in the file you want to edit.

  2. Add this to your agent's MCP config file. Replace args with the absolute path to mcp-server.js on your computer (on Windows, JSON strings need \\ for each backslash). FIGMA_TOKEN is optional — only needed if you want the REST API tools (get_figma_data, download_figma_images); omit that line entirely if you don't have a token yet:

{
  "mcpServers": {
    "figma-write-bridge": {
      "command": "node",
      "args": [
        "C:\\path\\to\\figma-write-bridge\\mcp-server.js"
      ],
      "env": {
        "FIGMA_BRIDGE_HOST": "127.0.0.1",
        "FIGMA_BRIDGE_PORT": "8787",
        "FIGMA_BRIDGE_CHANNEL": "default",
        "FIGMA_BRIDGE_TIMEOUT_MS": "180000",
        "FIGMA_TOKEN": "your_figma_personal_access_token"
      }
    }
  }
}

Notes:

  • If your agent starts the MCP server automatically, do not also run npm start (only one process can use port 8787).

  • After adding the config, restart your AI agent app so it picks up the new server.

  • Running more than one MCP server? Each server needs its own port and channel: set FIGMA_BRIDGE_PORT and FIGMA_BRIDGE_CHANNEL per server (e.g. server A → port 8787, channel default; server B → port 8788, channel design). Each server should use a port in the plugin's scan range (8787–8797) so it shows up in the plugin's Discovered servers dropdown — then in Figma just pick the server for the agent you want. One plugin UI connects to exactly one channel / MCP server.

  • MCP tools let the agent call Figma. The skill in the next section tells it how to use those tools — zip and upload it too.

Upload the Agent Skill (Required)

The MCP config is not enough on its own. Zip the skill folder and upload that zip in your AI agent (Cursor, Claude, etc.) so the agent loads Figma write-bridge rules.

  1. Zip skills/figma-write-bridge/ — the folder that contains SKILL.md, not the repo root and not the parent skills/ directory.

  2. Confirm the zip root is SKILL.md (open the zip: you should see SKILL.md immediately, not figma-write-bridge/SKILL.md).

  3. In your AI agent, upload / import that .zip as a skill (the agent's "Skills" or "Upload skill" UI).

  4. Restart or reload the agent if it does not pick up new skills automatically.

PowerShell (from the repo root):

Compress-Archive -Path "skills\figma-write-bridge\*" -DestinationPath "figma-write-bridge-skill.zip" -Force

macOS / Linux:

(cd skills/figma-write-bridge && zip -r ../../figma-write-bridge-skill.zip .)

The folder looks like this:

skills/figma-write-bridge/
├── SKILL.md          # always-on rules (~90 lines)
├── setup.md          # preconditions and channels
├── library.md        # file-library catalog
├── layout.md         # auto layout, hug/fill, placement, default fills
├── schema.md         # closed schema + error table
├── handoff.md        # screen annotations
├── tools.md          # full tool catalog
├── heuristics.md     # product-designer rules
└── playbooks.md      # copy / screen / component / prototype / motion

SKILL.md links those files one level deep; the agent should read only the file the current step needs. Re-zip and re-upload after you pull skill updates.

Target Frames (Safety)

set_target_frame / get_target_frames / clear_target_frames let the agent record which frame(s) you intend it to work in. Target-frame scoping is enforced by the plugin: when target frame(s) are set, write/delete actions targeting nodes outside those frame(s) are rejected with an error (create actions use the target frame as their host when one is set). A small, explicit set of delete/reset/clear actions is allowed by default (deleting a node, a page, a variable, a variable mode, or a component property/slot, and clearing prototype reactions) so the agent can actually make the changes you ask for — everything else matching "delete/remove/reset/clear" is blocked.

The server syncs targetFrameIds into the plugin automatically whenever they change (set_target_frame / clear_target_frames) and whenever a plugin connects, so enforcement stays in sync even if the plugin UI reloads.

Recommended workflow:

  1. In Figma, select the frame you want the AI to work on.

  2. From your AI agent, call get_selection and take the selected frame id.

  3. Call set_target_frame with that frameId.

  4. Use create/edit tools to add content within that frame.

Tip: If you call create_frame (or figma_create_frame) with no target set, the created frame becomes the target automatically.


Figma REST API Tools (No Plugin Required)

If you provide FIGMA_TOKEN, the server also exposes tools that call the Figma REST API directly:

  • get_figma_data (fetch file JSON, and optionally node JSON)

  • download_figma_images (download images/exports to a local folder)

  • list_comments / post_comment / delete_comment (file comments)

  • export_frames_to_disk (bulk-export a set of frames or a whole page to a local folder)

  • search_components (find components/component-sets across your account or a team)

Examples:

  • get_figma_data({ fileKey, nodeId? })

  • download_figma_images({ fileKey, nodes, localPath, pngScale? })

  • export_frames_to_disk({ fileKey, pageId, localPath, format: "png", scale: 2 })

  • post_comment({ fileKey, message, nodeId })

  • search_components({ teamId?, fileKey?, pageSize? })


Connect from Figma (Setup in Your File)

Use the Figma Desktop app. The plugin connects to the bridge over ws://localhost, which only works in the Figma Desktop app — the browser version cannot reach a local WebSocket server.

  1. Open the Figma file you want to work in.

  2. Run the plugin:

    • Plugins → Development → Figma Write Bridge (Local)

  3. The plugin auto-connects using the defaults shown (Server localhost:8787 — host and port in one field; ws:// is added automatically, Channel default) — if the server is already running, the status flips to Connected with nothing else to click.

  4. Discovered servers — on load (and via the Scan button) the plugin scans localhost ports 8787–8797 for running figma-write-bridge servers. Every running agent's MCP server appears in the dropdown, labelled channel · host:port — fileName. Picking one auto-fills Server + Channel and connects immediately, so with several agents you just choose which one should control this file.

  5. Channel defaults to default. If you started the MCP server with a non-default FIGMA_BRIDGE_CHANNEL (see Channels below), type that same channel here so the plugin joins the right server. Only touch the fields if you changed FIGMA_BRIDGE_PORT or use a custom channel, then click Connect.

As long as the plugin stays open and connected, the bridge can send commands into this Figma file. If you close and reopen the plugin, it reconnects automatically the same way.


Channels (One Plugin = One Channel / MCP Server)

A “channel” is just a name that targets the right Figma file through the right MCP server.

  • By default the channel is default, and that is the only channel a single server needs.

  • To run multiple MCP servers (e.g. one per Figma file/team), give each server its own port and channel:

    • Server A → FIGMA_BRIDGE_PORT=8787, FIGMA_BRIDGE_CHANNEL=default

    • Server B → FIGMA_BRIDGE_PORT=8788, FIGMA_BRIDGE_CHANNEL=design (or any custom name)

Then in each open Figma file, run the plugin and enter the matching server host:port and channel in the UI. Each plugin UI connects to exactly one channel / MCP server.

Your external tool can see and select channels per server:

  • figma_bridge_status lists every connected channel with its fileKey/fileName on that server's WebSocket.

  • join_channel switches which connected channel subsequent commands target.

  • list_channels shows the dashboard of connected channels.


Useful Notes / Safety

  • The bridge is meant to run locally. By default it binds to 127.0.0.1 (only your computer can access it).

  • Treat this like “edit access”: only run the bridge when you trust the tool/script driving it.

  • Keep a backup: duplicate your Figma file before running large generations/changes.

  • The plugin must remain open; if you close the plugin UI, the connection is lost.

  • Deleting top-level content requires confirmation: delete_node / delete_multiple_nodes refuse to remove a page, a top-level frame, or a top-level section unless you pass confirmFrameOrPageDeletion: true — an explicit safety guard against wiping a whole page/frame in one call.


Troubleshooting

  • Plugin says “Reconnecting…”

    • Make sure npm start is running and no firewall is blocking port 8787.

    • Confirm the WS URL matches the server (ws://localhost:8787). Both localhost:8787 and ws://localhost:8787 are accepted — the plugin adds the ws:// prefix if it’s missing.

    • Run the plugin in the Figma Desktop app (not a browser tab) — the browser version cannot connect to a local WebSocket server.

    • If you changed the plugin code, re-import the plugin from the manifest (Plugins → Development → Import plugin from manifest…) — Figma caches the previously imported copy and does not pick up file changes automatically.

  • Server/tool says “Figma plugin not connected”

    • Open the Figma file and run the plugin, then click Connect.

    • If you have multiple files, ensure you’re using the correct Channel.

  • “No servers found on ports 8787–8797”

    • The dropdown only lists servers running inside the scan range. Make sure npm start (or your agent) is actually running, and that each server uses a FIGMA_BRIDGE_PORT in 8787–8797. Otherwise type the host:port manually in Server.

    • Check a server directly with curl http://127.0.0.1:8787/health.

  • Port already in use

    • Start the server on a different port by setting FIGMA_BRIDGE_PORT, then use the same port in the plugin UI WS URL.


Advanced (Optional): Server Settings

Environment variables supported by the server:

  • FIGMA_BRIDGE_HOST (default 127.0.0.1)

  • FIGMA_BRIDGE_PORT (default 8787)

  • FIGMA_BRIDGE_CHANNEL (default default; pin this to run multiple MCP servers, each on its own port)

  • FIGMA_BRIDGE_TIMEOUT_MS (default 180000)

  • FIGMA_BRIDGE_MAX_RESULT_BYTES (default 50000) — cap on a single tool result before it is truncated, to stop one big read from filling the agent's context. Raise it if you genuinely need a large single read. Catalog tools (list_variables, list_variable_collections, export_tokens, get_styles, get_local_components) are exempt from the cap so the agent always receives the full catalog for that call — they are instead paged (limit/offset, default 500, plus total) so a single tool call stays token-cheap while every entry stays reachable. Truncation is JSON-safe: for JSON results it trims the largest arrays and sets a truncated: true flag instead of corrupting the payload.

  • FIGMA_TOKEN (required for the REST API tools: get_figma_data, download_figma_images, comments, export_frames_to_disk, search_components)

Example (PowerShell):

$env:FIGMA_BRIDGE_PORT="8790"
$env:FIGMA_BRIDGE_CHANNEL="default"
npm start

Then connect in Figma to ws://localhost:8790 with Channel default.

Port already in use? The server stops immediately with a clear message instead of silently failing. If you hit it, another figma-write-bridge instance is already bound to that port — stop it, use FIGMA_BRIDGE_PORT for a different one, and point the plugin UI at that new port (and matching channel).

Health check / discovery — every server answers GET http://127.0.0.1:<port>/health on its WebSocket port with { "name": "figma-write-bridge", "wsUrl", "host", "port", "channel", "connectedChannels" }. The Figma plugin uses this to populate the Discovered servers dropdown (scan range 8787–8797, tunable via the scanPortStart/scanPortCount constants at the top of figma-plugin/code.js). You can also check a server yourself, e.g. curl http://127.0.0.1:8787/health.

Available Tools

188 tools
add_component_propertyC

Add a BOOLEAN, TEXT, INSTANCE_SWAP, or VARIANT property to a component or component set.

Inputs: name (required), type (required), nodeId, componentId, +4 more.

append_to_slotB

Reparent existing nodes into a SLOT. Children join the slot's layout flow (AUTO + FILL when the slot is auto layout) at 0,0 when it is freeform. Do not move_node afterwards. Optionally pass index and layoutSizing.

Inputs: nodeIds (required), slotNodeId (required), index, ignoreAutoLayout, +2 more.

apply_animation_styleA

Apply a reusable animation style to a node. Identify it by styleId from list_animation_styles, or by styleName for a substring match. duration and timelineOffset are seconds and are top-level, not props. Returns the applied instance id needed by remove_animation_style. Verify the result against animationStyles in the response — applying a style does not materialize tracks into animations.

Inputs: nodeId (required), name, props, styleId, +5 more.

apply_effect_styleC

Apply an effect style to a node's effects.

Inputs: nodeId (required), styleId (required).

apply_fill_styleC

Apply a paint style to a node's fills.

Inputs: nodeId (required), styleId (required).

apply_grid_styleC

Apply a grid style to a frame's layout grids.

Inputs: nodeId (required), styleId (required).

apply_shaderA

Import (if needed) and apply a shader to a node. Effect shaders go on effects; fill shaders go on fills or strokes. properties may be keyed by property-definition id or by the author-defined name from propertyDefinitions. Prefer this over set_effects when applying a SHADER. Pass replace:false to append an effect shader alongside existing effects.

Inputs: nodeId (required), name, target, replace, +4 more.

apply_stroke_styleA

Apply a paint style to a node's strokes (design-system preferred). For raw color/weight/align/dash without a style, use set_stroke_color instead.

Inputs: nodeId (required), styleId (required).

apply_text_styleC

Apply a text style to a TEXT node.

Inputs: nodeId (required), styleId (required).

arrange_childrenC

Distribute the direct children of a frame/node along the main axis (horizontal or vertical). Accepts the same mode/gap/crossAlign options as distribute_nodes. Bounds default to the parent.

Inputs: parentNodeId (required), gap, axis, mode, +1 more.

bind_color_variable_to_fillC

Bind a COLOR variable to a node's fill paint.

Inputs: nodeId (required), variableId (required), paintIndex.

bind_color_variable_to_strokeC

Bind a COLOR variable to a node's stroke paint.

Inputs: nodeId (required), variableId (required), paintIndex.

bind_component_propertyC

Bind a BOOLEAN/TEXT/INSTANCE_SWAP property to a node field using componentPropertyReferences.

Inputs: field (required), nodeId (required), unbind, propertyName, +1 more.

bind_variable_to_propertyB

Bind a FLOAT/STRING/BOOLEAN/EASING/TIMING variable to a node property via setBoundVariable(property, variable). Use property "opacity" to token-bind layer opacity (pair a FLOAT variable with a color rather than detaching the fill). Also used for padding, radius, spacing, and fontSize.

Inputs: nodeId (required), property (required), variableId (required).

boolean_groupC

Combine 2+ vector nodes into a boolean group (UNION, SUBTRACT, INTERSECT, EXCLUDE).

Inputs: nodeIds (required), op, name, parentNodeId.

bring_to_frontA

Bring a layer to the front of its parent (z-order). In Figma, children[last] is front-most.

Inputs: nodeId (required).

bulk_renameB

Find-and-replace text in node names across a subtree (defaults to current page). Supports regex and dryRun. Returns a before/after diff.

Inputs: find (required), dryRun, replace, useRegex, +1 more.

bulk_updateB

Apply one property across many nodes. Supported properties: fillColor, cornerRadius, opacity, visible, name, fillStyle, textStyle, cornerRadii. Target by nodeIds, by nodeTypes under rootNodeId, or the whole current page.

Inputs: property (required), value, nodeIds, nodeTypes, +1 more.

clear_reactionsC

Remove all prototype reactions from a node.

Inputs: nodeId (required).

clear_target_framesB

Clears the active target frameIds.

Inputs: no named parameters.

clone_nodeC

Create a copy of an existing node with optional position offset and name.

Inputs: nodeId (required), dx, dy, name.

clone_node_into_parentA

Clone a node and append it into a specified parent container (frame, section, group, auto-layout, slot, or page). In auto-layout parents the copy joins the flow unless dx/dy are non-zero (then it is set to absolute positioning).

Inputs: nodeId (required), parentNodeId (required), dx, dy, +2 more.

combine_as_variantsC

Combine existing component nodes into a component set and lay them out to avoid overlap.

Inputs: componentIds (required), gap, gapX, gapY, +4 more.

create_checkpointA

Snapshot a handful of common mutable properties (position, size, rotation, opacity, visibility, fills, strokes, corner radius, text characters) on the given nodes so they can be restored later with restore_checkpoint. NOT true undo: it cannot restore a deleted node or undo structural changes (reparenting, new/removed children), and state is lost if the Figma plugin UI reloads.

Inputs: nodeIds (required), label.

create_componentA

Create a new empty component. Pass parentNodeId to insert into a frame, auto-layout stack, or slot. Inside auto layout omit x/y and use index. Nested into auto layout, omit width/height unless you want a fixed size.

Inputs: x, y, name, index, +6 more.

create_component_from_nodeC

Convert an existing node into a main component.

Inputs: nodeId (required), name.

create_component_instanceA

Create an instance of a local component. Pass parentNodeId to insert into a frame, auto-layout stack, or slot. Inside auto layout omit x/y and use index.

Inputs: componentId (required), x, y, index, +4 more.

create_component_slotC

Create a slot inside a component variant. This also creates the corresponding SLOT property.

Inputs: x, y, name, width, +5 more.

create_effect_styleC

Create or update a local effect style.

Inputs: name (required), effects (required).

create_frameA

Create a new frame. Defaults to NO fill (transparent) — Figma's native white fill is cleared so layout containers stay invisible. Pass fillHex only when the frame is a visible surface (card, screen bg, chip). Nested into auto layout: omit x/y and omit width/height (do not ship at 320×200); pass parentNodeId + index + layoutSizing. Optional layoutMode (HORIZONTAL|VERTICAL|GRID).

Inputs: b, g, r, x, +13 more.

create_grid_styleC

Create or update a local grid style.

Inputs: name (required), layoutGrids (required).

create_instance_from_component_keyB

Create an instance from a library component key inside the target frame/parent. Pass parentNodeId + index for auto-layout stacks; omit x/y unless the parent is freeform.

Inputs: componentKey (required), x, y, index, +4 more.

create_instance_from_instanceA

Create a new instance of whatever main component an existing instance points at. Pass parentNodeId (or parentId) to drop it into a frame, stack, or slot. Inside auto layout omit x/y and use index.

Inputs: instanceId (required), x, y, index, +5 more.

create_instance_from_set_keyA

Create an instance from a library component set key (default variant) inside the target frame/parent. Pass parentNodeId + index for auto-layout stacks; omit x/y unless the parent is freeform.

Inputs: componentSetKey (required), x, y, index, +4 more.

create_pageC

Create a new page in the document. Pass activate: true to also switch to it.

Inputs: name, activate.

create_paint_styleC

Create or update a local paint style.

Inputs: name (required), hex, paints.

create_rectangleA

Create a rectangle. Pass parentNodeId to insert into a frame, auto-layout stack, or slot. Inside auto layout, omit x/y and use index + layoutSizing; x/y only apply to freeform frames or ignoreAutoLayout overlays.

Inputs: x, y, name, index, +6 more.

create_sectionC

Create a SECTION node. Optional fillColor {r,g,b,a}, sectionProperties (e.g. {sectionType}), and parentNodeId.

Inputs: x, y, name, width, +4 more.

create_textA

Create a text node. Defaults to auto-width hug (textAutoResize WIDTH_AND_HEIGHT + HUG sizing). Pass parentNodeId to insert into a frame/stack/slot — inside auto layout omit x/y and use index. Pass textAlignHorizontal (LEFT|CENTER|RIGHT|JUSTIFIED), textAlignVertical (TOP|CENTER|BOTTOM), textAutoResize (WIDTH_AND_HEIGHT|HEIGHT|NONE|TRUNCATE), and layoutSizingHorizontal/Vertical (FIXED|HUG|FILL). For variable fonts, pass variationSettings (e.g. {wght:550}) and optionally omit fontStyle.

Inputs: characters (required), b, g, r, +16 more.

create_text_styleB

Create or update a local text style. For variable fonts, pass variationSettings (OpenType axis map) and optionally omit fontStyle so Figma picks the matching named instance.

Inputs: name (required), fontFamily (required), fills, fillsHex, +8 more.

create_typography_scaleB

Create a text-style scale (caption/body/h3/h2/h1/display by default, or custom steps) from a baseSize and ratio: fontSize = base * ratio^offset. Can also create sample text nodes in a frame (createSampleFrame) spanning the steps. For variable fonts, pass variationSettings and optionally omit fontStyle.

Inputs: ratio, steps, prefix, baseSize, +8 more.

create_variableB

Create a local variable in a collection and set values by mode. resolvedType: COLOR, FLOAT, STRING, BOOLEAN, EASING, or TIMING — EASING and TIMING hold motion easing curves and durations, letting a design system tokenize animation alongside color and spacing.

Inputs: name (required), collectionId (required), resolvedType (required), scopes, +3 more.

create_variable_collectionC

Create a local variable collection, optionally naming modes.

Inputs: name (required), modes.

create_variable_modeC

Add a new mode to a variable collection.

Inputs: name (required), collectionId (required).

create_vectorC

Create a VECTOR node from SVG path data. vectorPaths: [{data, windingRule?}]. Supports fills, strokes, strokeWeight, parentNodeId, x, y.

Inputs: vectorPaths (required), x, y, name, +4 more.

delete_commentB

Delete a comment from a Figma file via the REST API (requires FIGMA_TOKEN).

Inputs: fileKey (required), commentId (required).

delete_component_propertyB

Delete an existing component property from a component or component set.

Inputs: propertyName (required), nodeId, componentId, componentSetId, +1 more.

delete_component_slotA

Remove a SLOT node from a component and delete its corresponding SLOT property. Any content placed in instances of that slot is discarded.

Inputs: slotNodeId (required).

delete_multiple_nodesA

Delete multiple nodes by nodeIds (only within the allowed target frame). When any nodeId is a page or top-level frame, confirmFrameOrPageDeletion: true is required for that node.

Inputs: nodeIds (required), confirmFrameOrPageDeletion.

delete_nodeA

Delete a node by nodeId (only within the allowed target frame). Deleting a page or a top-level frame requires confirmFrameOrPageDeletion: true.

Inputs: nodeId (required), confirmFrameOrPageDeletion.

delete_pageA

Delete a page by pageId. Requires confirmDelete=true.

Inputs: pageId (required), confirmDelete (required).

delete_variableB

Delete an existing variable by variableId. Requires confirmDelete=true.

Inputs: variableId (required), confirmDelete (required).

delete_variable_collectionA

Delete an entire variable collection (and all its variables/modes). Requires confirmDelete=true.

Inputs: collectionId (required), confirmDelete (required).

delete_variable_modeC

Remove a mode from a variable collection. Requires confirmDelete=true.

Inputs: modeId (required), collectionId (required), confirmDelete (required).

distribute_nodesA

Space or align a set of nodes along an axis. axis: horizontal|vertical. mode: gap (fixed gap), spaceBetween/evenly (fill bounds), center (center the cluster). crossAlign: none|start|center|end. Bounds default to the common parent; pass bounds {x1,y1,x2,y2} to override (horizontal coordinates) or {y1,x1,y2,x2} semantics for vertical.

Inputs: nodeIds (required), gap, axis, mode, +2 more.

download_figma_imagesC

Download SVG/PNG/GIF images used in a Figma file via the Figma REST API.

Inputs: nodes (required), fileKey (required), localPath (required), pngScale.

duplicate_pageA

Duplicate a page (including all contents) by pageId. Auto-suffixes the name (Page 2, Page 3...) unless a name is given. Pass activate: true to switch to the duplicate.

Inputs: pageId (required), name, activate.

edit_component_propertyC

Rename or update the default value/preferred values of an existing component property.

Inputs: updates (required), propertyName (required), nodeId, componentId, +2 more.

edit_component_slotB

Rename, resize, or reposition an existing SLOT node inside a component. Figma keeps the SLOT property's name in sync with the slot node's name.

Inputs: slotNodeId (required), x, y, name, +2 more.

export_frames_to_diskA

Bulk-export frames from a Figma file to local disk via the REST API (requires FIGMA_TOKEN). Pass nodeIds, or pass pageId to export all top-level frames on a page. Renders PNG/JPG/SVG/PDF into a folder inside the figma-write-bridge repo.

Inputs: fileKey (required), localPath (required), scale, format, +3 more.

export_node_as_imageA

Export a node as an image (PNG, JPG, SVG, or PDF) or as video (MP4, GIF, WEBM) for a top-level frame with Motion. Prefer localPath so bytes are written to disk inside the figma-write-bridge repo instead of returning base64. If the payload would exceed the MCP result cap, the server auto-saves under exports/ and returns the path. Video export is only available in Figma Desktop for animated top-level frames.

Inputs: nodeId (required), fps, scale, format, +3 more.

export_tokensExport design tokensA

Export local Figma variables as a W3C-style Design Tokens object (nested by collection/variable name), plus a flat variables list. Colors are emitted as hex. With includeModes (default true) every mode's values are returned under tokensByMode like "Color/Dark"; the response also includes collections (each with its full modes list), modes (all mode keys emitted) and modeCount so you can verify all modes (not just the first) came through. Set includeModes=false to skip per-mode views. Pass collections (array of collection names or ids) to export only those collections — keeps the response small on large files.

Inputs: collections, includeModes.

extract_component_setB

Convert multiple existing frames (or components) into a variant component set: each frame becomes a COMPONENT, then they are combined into a COMPONENT_SET via combine_as_variants. Set propertyName to attempt adding a VARIANT property (best-effort).

Inputs: nodeIds (required), gap, gapX, gapY, +4 more.

figma_bridge_statusA

Returns whether the local Figma plugin is connected, the active channel, and every connected channel with the Figma file it belongs to. The channel defaults to "default" and is tied to this server's FIGMA_BRIDGE_CHANNEL; the plugin UI joins that channel (one plugin = one channel/server).

Inputs: no named parameters.

figma_set_solid_fillSet solid fillC

Sets a SOLID fill on a node by nodeId.

Inputs: b (required), g (required), r (required), nodeId (required), +1 more.

figma_set_textSet textB

Sets characters on a TEXT node (by nodeId or current selection).

Inputs: characters (required), nodeId.

find_and_replace_textA

Search TEXT node characters for a literal string or regex and replace matches, optionally across every page in the file (not just the current one). Pass dryRun: true first to preview matches before committing.

Inputs: query (required), replacement (required), dryRun, allPages, +5 more.

find_nodesA

Query the document for nodes matching a set of predicates, evaluated inside Figma so only matching rows come back (use this instead of reading a whole subtree and filtering). All predicates are optional and are ANDed together. Returns a columnar table: fields names the columns, rows holds one array per match, plus scanned/total/truncated counts. Examples: {types:["INSTANCE"], mainComponentName:"Button", fillHex:"#ff0000"} finds red button instances; {missingFillStyle:true} finds hardcoded fills with no style or variable bound (design-system drift); {types:["INSTANCE"], hasOverrides:true} finds overridden instances.

Inputs: name, limit, types, fields, +18 more.

generate_gridA

Generate a grid of items inside a parent. Clones itemNodeId if given, otherwise creates rectangles of itemWidth x itemHeight. Use {i} in name for the running index.

Inputs: name, rows, columns, spacingX, +5 more.

generate_paletteB

Generate a tonal 50..900 palette (default 10 steps) from a seed hex color. Light steps mix toward white, dark steps toward black. Optionally creates paint styles (createStyles=true), COLOR variables in a ' Tokens' collection (createVariables=true), and a swatch frame with labeled rectangles (createFrame=true).

Inputs: hex (required), gap, name, steps, +7 more.

get_all_pagesA

Compact map of every page in the open file: id/name/childCount. Pass includeTopLevel: true to also list each page's top-level frames (id/name/type/childCount). Use this once to get a full-file overview before targeted reads.

Inputs: includeTopLevel.

get_animation_presetsA

List curated motion presets plus the official Figma transition and easing values supported by this bridge. Use these values with set_reactions, upsert_reaction, set_transition_reaction, and set_smart_animate_reaction.

Inputs: no named parameters.

get_annotationsA

Get annotations on a node or in a subtree. Page-wide scans are paged (default limit 100). Pass nodeId for one node, or rootNodeId to scope. includeCategories defaults to false (pass true to expand category labels).

Inputs: limit, nodeId, offset, rootNodeId, +1 more.

get_changes_sinceB

Return nodes this bridge mutated since a cursor (re-read only what changed, not the whole doc). Pass prior currentSeq as sinceSeq to page. Cursor resets when the MCP server restarts.

Inputs: sinceSeq.

get_component_property_definitionsC

Inspect component or component-set property definitions for authoring and verification.

Inputs: nodeId, componentId, componentSetId, preferComponentSet.

get_document_infoC

Get information about the current Figma document.

Inputs: no named parameters.

get_document_treeA

Compact structural tree of the whole document (or a subtree). Returns a flattened columnar table: fields lists the column names, rows holds one array per node in pre-order, and the depth column gives nesting level (depth 0 is the root, each +1 is a child of the nearest preceding row with depth-1). Every node is {id, name, type}; no extra fields unless requested. Default maxDepth is 3. Hidden layers are omitted unless includeHidden is true. Pass fields to pull extra per-node values (characters, fills, fillHex, fillCount, strokes, x, y, width, height, absoluteBoundingBox, strokeWeight, cornerRadius, fillStyleId, strokeStyleId, textStyleId, layoutMode, layoutWrap, layoutSizingHorizontal, layoutSizingVertical, layoutPositioning, primaryAxisAlignItems, counterAxisAlignItems, layoutGrow, itemSpacing, padding, visible, opacity). Use rootNodeId to scope to a frame/page and excludeTypes (e.g. ["VECTOR"]) to drop icon noise. verbose returns the nested tree instead.

Inputs: fields, verbose, maxDepth, rootNodeId, +2 more.

get_eventsA

Read pushed Figma events (selectionchange/documentchange) since a sequence cursor. Pass the previous call's currentSeq as sinceSeq to page forward. Cursor state lives in the MCP server process and resets on restart.

Inputs: sinceSeq.

get_figma_dataA

Get Figma file data via the REST API. Pass nodeId to fetch only that node (avoids pulling the whole file). depth limits recursion on file/nodes. Returns file (or nodes), plus styles/components/component_sets when no nodeId is given.

Inputs: fileKey (required), depth, nodeId.

get_font_listA

List the distinct fonts (family + style, plus variationSettings when the text uses a variable font) used in the current page or a rootNodeId subtree, with usage counts. Pair with get_font_variation_axes to inspect a family's OpenType axes.

Inputs: rootNodeId.

get_font_variation_axesA

Return the OpenType variation axes a font family exposes (e.g. wght, slnt, opsz), or null/variable:false for a static family. Use this before setting variationSettings on create_text / set_text_style / create_text_style. Requires Figma Desktop with Plugin API Update 138 (variable fonts).

Inputs: family (required).

get_grid_layoutA

Read a GRID frame's track counts, gaps, per-track sizes, and every child's cell, span, and alignment. Returns isGrid:false for a frame that is not in GRID layout. Call this before repositioning children so you know the grid's bounds.

Inputs: nodeId (required).

get_instance_propertiesC

Get componentProperties for an instance.

Inputs: instanceId (required).

get_instance_slotsB

List SLOT nodes inside an instance (for inserting content into slots).

Inputs: instanceId (required).

get_instance_sourceB

Get main component/component-set keys and properties for an instance (to verify design system provenance).

Inputs: instanceId (required).

get_local_componentsA

Get information about local components across all pages: id, name, type, description, publish key, and a property count, returned as { components, total, offset, limit, pageCount } so you can page every component (default limit 500). Pass includeProperties: true to also get simplified component property definitions (for building library catalogs). Pass verbose to skip compacting; the columnar format is used otherwise.

Inputs: limit, offset, verbose, includeProperties, +1 more.

get_motionA

Read Motion state — timelines (with durations in seconds), the current playheadPosition in seconds (undefined when no timeline is active), manual keyframe tracks, applied animation styles, and resolved animations — for the given nodes, or the current selection when none are given. Motion is distinct from prototype reactions: reactions link frames on click, Motion animates properties over a timeline inside one top-level frame. Returns motionEnabled:false when the account lacks the Motion feature flag.

Inputs: nodeId, nodeIds.

get_node_infoA

Compact layout-aware summary of a node: size, auto-layout (mode, hug/fill, positioning, padding, gap, alignment), fillHex/fillCount, truncated text, and childCount. VECTOR children are omitted unless excludeTypes is []. Default maxDepth is 0 (the node itself). Pass fields to request a subset, or verbose for the raw REST dump.

Inputs: nodeId (required), fields, verbose, maxDepth, +1 more.

get_nodes_infoA

Compact layout-aware summaries for several nodes. Same defaults as get_node_info. Prefer this over looping get_node_info.

Inputs: nodeIds (required), fields, verbose, maxDepth, +1 more.

get_overlay_settingsA

Read a frame/component's overlay prototype settings (position type, background, click-outside behavior). These control HOW an overlay appears; the interaction that opens it is a NODE reaction with navigation OVERLAY, set via set_reactions or set_transition_reaction.

Inputs: nodeId (required).

get_parent_chainC

Walk up the parent chain of a node, returning id/name/type at each level.

Inputs: nodeId (required), maxDepth, stopAtId.

get_prototype_settingsB

Get the current page's prototype start node and Flows starting points.

Inputs: no named parameters.

get_reactionsC

Get all prototype reactions from nodes.

Inputs: nodeIds (required).

get_selectionB

Get information about the current selection.

Inputs: no named parameters.

get_selection_contextA

One-call bundle for the current selection: compact layout-aware node info plus (for instances) main component id, property definitions/values, and slots. Use instead of chaining get_selection→get_node_info→get_component_property_definitions.

Inputs: fields, verbose, maxDepth, excludeTypes.

get_style_guideA

Extract a usage style guide from the current page (or rootNodeId subtree): counts of distinct solid colors (hex), color variable bindings, font family/style combos, font sizes, line heights, spacing/gap/padding values, corner radii, stroke weights, and opacities.

Inputs: rootNodeId.

get_stylesA

Get information about local styles. Paged per style type: pass limit/offset to bound each response (default 500); totalPaintStyles/totalTextStyles/totalEffectStyles/totalGridStyles report the full counts.

Inputs: limit, offset.

get_target_framesA

Returns the current target frameIds the agent is allowed to modify.

Inputs: no named parameters.

get_variableA

Read one variable's values in EVERY mode. Find it by variableId, by publish key, or by name (if the name is not unique, pass collectionId/collectionName to disambiguate). Returns valuesByMode / valuesByModeName (raw values, aliases as {type:"VARIABLE_ALIAS",id}) plus resolvedValuesByMode / resolvedValuesByModeName that follow aliases to a concrete value (colors as hex) with cycle/unresolved markers if needed.

Inputs: key, name, variableId, collectionId, +1 more.

group_nodesC

Wrap existing nodes in a GROUP.

Inputs: nodeIds (required), name, parentNodeId.

import_component_by_keyC

Import a library component into the current file using its componentKey. Optionally rename the imported main component.

Inputs: componentKey (required), name.

import_component_set_by_keyA

Import a library component set into the current file using its componentSetKey. Optionally rename the imported main component set.

Inputs: componentSetKey (required), name.

import_shader_by_idA

Import a shader into the current file so it can be applied. Pass shaderId (from list_shaders) or name. Idempotent if the shader is already imported; returns the shader with imported:true and populated propertyDefinitions.

Inputs: name, shaderId.

import_style_by_keyC

Import a published library style into the file by key.

Inputs: key (required).

import_tokensImport design tokensA

Import a W3C-style Design Tokens JSON object into Figma variables (and optionally paint styles). Accepts nested {group:{name:{$type,$value}}} or plain nested values (types inferred from values). Creates/updates a variable collection (default 'Design Tokens') and a Default mode, then sets values. color -> COLOR variable + paint style, number/dimension -> FLOAT, string -> STRING, boolean -> BOOLEAN.

Inputs: tokens (required), modeName, createStyles, collectionName.

import_variable_by_keyC

Import a published library variable into the file by key.

Inputs: key (required).

insert_childA

Insert an existing child node at an index inside a parent container. Omit index to append at the end. Inside auto layout this keeps the child in the flow (layoutPositioning AUTO) unless ignoreAutoLayout is true.

Inputs: childId (required), parentId (required), index, ignoreAutoLayout, +2 more.

join_channelA

Selects which connected Figma plugin channel to target for subsequent commands. Channels default to "default" and are configured per MCP server via FIGMA_BRIDGE_CHANNEL; the plugin UI joins that channel. Use figma_bridge_status to see which channel maps to which file.

Inputs: channel (required).

list_animation_stylesA

List Figma's first-party animation styles, each with its styleId and the props it accepts. Call this before apply_animation_style to get a real styleId. Note name is an i18n key (e.g. 'motion.preset_name.position') and each entry in props is a documentation string describing the prop, not a value to send back.

Inputs: no named parameters.

list_channelsA

Channel dashboard: lists every connected Figma plugin channel with its fileKey/fileName and connection time.

Inputs: no named parameters.

list_checkpointsB

List checkpoints captured so far in this plugin session.

Inputs: no named parameters.

list_commentsC

List comments on a Figma file via the REST API (requires FIGMA_TOKEN).

Inputs: fileKey (required).

list_shadersA

List shader effects and fills available to this file (in-file, subscribed libraries, and owned shaders), with ids, type (effect|fill), and imported. Property definitions are omitted unless includeProperties is true — import_shader_by_id already returns them. Shaders with imported:false must be imported before apply_shader.

Inputs: includeProperties.

list_variable_collectionsB

List local variable collections in the current file.

Inputs: no named parameters.

list_variablesA

List local variables in the current file (optionally filtered by resolvedType). Pass includeScopes: true to also return each variable's scopes, or includeValues: true to also return each variable's value in EVERY mode (valuesByMode by modeId, valuesByModeName by mode name, defaultValue from the collection's first mode, and the mode list) — use this to read theme/dark-mode values, not just the default mode. Results are paged: default limit 500 per call with total/offset/pageCount so you can page through every variable instead of reading them all at once.

Inputs: limit, offset, resolvedType, includeScopes, +1 more.

move_component_to_fileMove component to another fileA

Import a component or component-set from this file into another connected channel's file, then optionally delete the source (move). Requires the target file to be open with the plugin connected to this server on a different channel (see figma_bridge_status / list_channels). Use mode: 'copy' to keep the source. Accepts componentId, componentKey, or componentName to identify the source.

Inputs: targetChannel (required), mode, name, componentId, +2 more.

move_nodeA

Move a node on the canvas. Pass absolute x/y, or relative dx/dy to nudge. Inside auto layout this is rejected unless ignoreAutoLayout is true (Ignore auto layout / overlay). Reorder stacks with insert_child; align with set_axis_align.

Inputs: nodeId (required), x, y, dx, +2 more.

move_node_to_pageB

Move (cut) a top-level node to another page, or copy it there (duplicate into page, keeping the original). Set copy: true to keep the original.

Inputs: nodeId (required), targetPageId (required), copy, name.

post_commentA

Create a comment on a Figma file via the REST API (requires FIGMA_TOKEN). Optional nodeId anchors it to a node; clientMeta {x,y[,nodeId]} positions it on the canvas.

Inputs: fileKey (required), message (required), nodeId, clientMeta.

read_my_designA

Compact layout-aware summary of the current selection. Default fields cover size, auto-layout (mode, hug/fill, padding, gap, alignment), fillHex/fillCount, and truncated text. VECTOR children are omitted unless you pass excludeTypes: []. Pass verbose for the raw REST dump. Use instead of chaining get_selection when you only need structure.

Inputs: fields, verbose, maxDepth, excludeTypes.

redoRedo last undone actionA

Re-applies the most recently undone action (snapshot-based, best-effort).

Inputs: no named parameters.

remove_animation_styleA

Remove one applied animation style from a node by its appliedId (read it from get_motion's animationStyles), or pass all:true to remove every applied style.

Inputs: nodeId (required), all, appliedId.

remove_keyframe_trackA

Remove one manual keyframe track from a node by field (same field naming as set_keyframe_track), or pass all:true to clear every manual track on the node.

Inputs: nodeId (required), all, field, paintIndex.

rename_nodeC

Renames a node by nodeId.

Inputs: name (required), nodeId (required).

rename_pageC

Rename a page by pageId.

Inputs: name (required), pageId (required).

rename_variableC

Rename an existing variable by variableId.

Inputs: name (required), variableId (required).

rename_variable_collectionC

Rename a variable collection by collectionId.

Inputs: name (required), collectionId (required).

rename_variable_modeC

Rename a mode within a variable collection.

Inputs: name (required), modeId (required), collectionId (required).

reorder_grid_tracksB

Move whole rows or columns of a GRID frame, taking their children with them. fromIndices lists the 0-based tracks to move and insertionIndex is where they land.

Inputs: axis (required), nodeId (required), fromIndices (required), insertionIndex (required).

reorder_pageB

Move a page to a new index in the page tab bar (0-based).

Inputs: index (required), pageId (required).

reparent_nodeB

Move (cut) an existing node into a new parent container (frame, section, group, auto-layout, slot, or page). Inside auto layout the node joins the flow (omit x/y, pass index). Pass ignoreAutoLayout + x/y only for overlays.

Inputs: nodeId (required), newParentId (required), x, y, +4 more.

replace_all_instancesA

Swap every instance whose main component key matches sourceComponentKey to targetComponentKey. Supports dryRun and a rootNodeId scope.

Inputs: sourceComponentKey (required), targetComponentKey (required), dryRun, rootNodeId.

resize_nodeA

Resize a node. Width and/or height may be passed. For TEXT: WIDTH_AND_HEIGHT (hug) refuses resize — pass textAutoResize HEIGHT (width only) or NONE (fixed). Setting only width on hug auto-promotes to HEIGHT. Prefer set_layout_sizing for HUG/FILL.

Inputs: nodeId (required), width, height, forceFixed, +1 more.

resize_to_fitA

Fit a layer. Two modes: (1) pass targetNodeId to scale nodeId to fit inside that layer, preserving aspect ratio and centering it (fit: 'contain' letterboxes, 'cover' fills and crops); (2) omit targetNodeId to shrink-wrap nodeId to tightly fit its own children (Figma's 'Resize to Fit').

Inputs: nodeId (required), fit, targetNodeId.

restore_checkpointC

Reapply a snapshot captured by create_checkpoint to whichever of its nodes still exist. See create_checkpoint for what is and isn't covered.

Inputs: checkpointId (required).

run_batchA

Execute multiple bridge actions in one round trip instead of one WebSocket call per action. Runs sequentially inside the plugin; by default stops at the first error (partial results are still returned in order). This is NOT a transaction: steps that already succeeded are not rolled back if a later step fails. Use create_checkpoint first if you need a rollback path for the nodes you're about to batch-edit.

Inputs: actions (required), stopOnError.

scan_instances_with_sourcesC

Scan instances under a root node and return their main component/component-set keys.

Inputs: offset, chunkSize, rootNodeId.

scan_nodes_by_typesA

List nodes of given types in a subtree. Prefer find_nodes for anything more selective than a type filter. Defaults to the current page, or the single target frame if one is set. Pass rootNodeId to scope. Paged (default limit 200). Returns a columnar table plus total/offset/limit/truncated.

Inputs: types (required), limit, offset, verbose, +1 more.

scan_text_nodesC

Scan text nodes with basic chunking support. Matches are returned as a columnar table: fields lists the column names and rows holds one array per node in the same order.

Inputs: offset, verbose, maxChars, chunkSize, +1 more.

search_componentsA

Search for components/component-sets via the Figma REST API (requires FIGMA_TOKEN). Provide teamId to use /v1/team/{teamId}/components, or omit it to use /v1/me/components. Optional fileKey and pageSize filter the results.

Inputs: type, teamId, fileKey, pageSize.

send_to_backA

Send a layer to the back of its parent (z-order). In Figma, children[0] is back-most.

Inputs: nodeId (required).

set_annotationC

Create or update an annotation with markdown support.

Inputs: nodeId (required), labelMarkdown (required), categoryId, properties.

set_auto_layoutB

Set multiple auto-layout properties in one call. sizing may include primaryAxisSizingMode/counterAxisSizingMode and/or layoutSizingHorizontal|width + layoutSizingVertical|height (FIXED|HUG|FILL). Empty frames keep Figma's default white fill stripped unless you pass clearFill:false.

Inputs: frameId (required), width, height, sizing, +9 more.

set_axis_alignA

Set primary and counter axis alignment for auto-layout frames. primaryAxisAlignItems: MIN, CENTER, MAX, SPACE_BETWEEN, SPACE_AROUND, or SPACE_EVENLY (SPACE_AROUND and SPACE_EVENLY distribute children with equal space around or between them, padding included). counterAxisAlignItems: MIN, CENTER, MAX, or BASELINE.

Inputs: nodeId (required), counterAxisAlignItems, primaryAxisAlignItems.

set_corner_radiusC

Set the corner radius of a node with optional per-corner control.

Inputs: nodeId (required), radius (required), corners.

set_current_pageC

Set the current/active page by pageId.

Inputs: pageId (required).

set_effectsA

Set effects on a node. Pass effectStyleId to apply an existing style, or effects as a raw array. Supported types: DROP_SHADOW and INNER_SHADOW ({color, offset:{x,y}, radius, spread}; DROP_SHADOW also takes showShadowBehindNode), LAYER_BLUR and BACKGROUND_BLUR ({radius, blurType:'NORMAL'} or blurType:'PROGRESSIVE' with startRadius/startOffset/endOffset), NOISE ({color, noiseSize, density, noiseType:'MONOTONE'|'DUOTONE'|'MULTITONE'}), TEXTURE ({noiseSize, radius, clipToShape}), GLASS ({lightIntensity, lightAngle, refraction, depth, dispersion, radius}), and SHADER ({id, properties}) after the shader has been imported. Prefer apply_shader for shaders. Figma requires visible and blendMode on shadows and visible plus blurType on blurs; the bridge fills those in and strips any key the effect variant does not declare, so pass only the fields you care about and never echo back an effect read off another node. boundVariables.color can bind a variable to the shadow color.

Inputs: no named parameters.

set_fill_colorB

Set or clear a node's solid fill. Prefer applying styles/variables when available. Pass clear:true (or fills:[]) to remove fills from layout containers that should be transparent. Also accepts fillHex.

Inputs: nodeId (required), b, g, r, +3 more.

set_fixed_childrenC

Mark direct children of a frame as fixed in a scrolling prototype (fix position when scrolling).

Inputs: frameId (required), fixedChildIds (required).

set_flow_starting_pointsB

Replace the current page's Flows list (named prototype entry points), each pointing at a top-level FRAME.

Inputs: flowStartingPoints (required).

set_focusA

Select a single node and scroll the viewport to it. Switches to the node's page first, so this works with ids returned by a cross-page find_nodes. Use this to show the user what you just built or found.

Inputs: nodeId (required).

set_gradient_fillB

Set a gradient fill (LINEAR, RADIAL, ANGULAR, DIAMOND) on a node. stops: [{position 0..1, color {r,g,b[,a]}}]. Optional from/to transform points (normalized), opacity, paintIndex.

Inputs: stops (required), nodeId (required), to, from, +3 more.

set_grid_child_positionA

Place a child in its GRID parent: row/column are 0-based anchor indices, rowSpan/columnSpan are how many tracks it covers, and the aligns control it inside its cell ('AUTO' stretches). Spans must fit inside the grid — anchor + span cannot exceed the track count, and the tool reports the actual bound if it does. A span also cannot cross a cell another child already occupies (Figma reports it as a span blocked by existing children in adjacent columns), so place and widen a spanning child BEFORE adding the children beside it. Children keep their own size by default; use align 'AUTO' on an axis to stretch into the cell.

Inputs: nodeId (required), row, column, rowSpan, +3 more.

set_grid_layoutA

Turn a frame into a GRID auto-layout container and configure it. GRID is a third layoutMode alongside HORIZONTAL/VERTICAL: children occupy cells rather than a single flow. Set rowCount/columnCount and the gaps, and optionally size individual tracks via rowSizes/columnSizes — each entry is a number (a FIXED px size) or {type:'FLEX'|'FIXED'|'HUG', value}. FLEX tracks share leftover space by their value as a weight. Position children afterwards with set_grid_child_position.

Inputs: nodeId (required), rowGap, rowCount, rowSizes, +5 more.

set_image_fillA

Set an IMAGE fill on a node from a URL (createImageAsync), raw base64 imageBytes, or a local image file via localPath (read as base64 server-side). scaleMode: FILL, FIT, CROP, TILE.

Inputs: no named parameters.

set_instance_propertiesC

Set component properties/variants on an instance.

Inputs: instanceId (required), properties (required).

set_item_spacingC

Set distance between children in an auto-layout frame.

Inputs: nodeId (required), itemSpacing (required).

set_keyframe_trackA

Add or replace one manual keyframe track on a node. field is a property name (TRANSLATION_X/Y/XY, ROTATION, SCALE_X/Y/XY, OPACITY, CORNER_RADIUS, STROKE_WEIGHT, WIDTH, HEIGHT, STACK_* / GRID_* spacing, PATH_TRIM_START/END) or FILLS/STROKES/EFFECTS with a paintIndex. Transform fields COMPOSE with the node's resting transform (neutral 0, or 1 for scale); the others replace the value. timelinePosition is in seconds; the first keyframe's value holds back to t=0 and the last holds to the end, so no padding keyframes are needed. Easing on a keyframe describes the move INTO it, and adds 'HOLD' for step interpolation. The containing timeline is extended when the animation runs past it unless extendTimeline is false. Animate descendants: a top-level frame owns the timeline rather than animating on it, so keyframes there do nothing and are refused unless allowTopLevelFrame is true. Color and effect animation uses field FILLS, STROKES, or EFFECTS together with paintIndex.

Inputs: field (required), nodeId (required), track, keyframes, +3 more.

set_layout_gridsA

Set layout grids (layout guides) on a frame. Each grid is either {pattern:'ROWS'|'COLUMNS', alignment:'MIN'|'MAX'|'CENTER'|'STRETCH', gutterSize, count, sectionSize?, offset?} or {pattern:'GRID', sectionSize}, plus optional visible and color. Figma rejects sectionSize when alignment is STRETCH and offset when alignment is CENTER; the bridge drops those rather than failing, and strips any key the pattern does not declare.

Inputs: frameId (required), layoutGrids (required).

set_layout_modeA

Set the layout mode and wrap behavior of a frame: NONE, HORIZONTAL, VERTICAL, or GRID. GRID is a full cell-based layout — configure its tracks with set_grid_layout and place children with set_grid_child_position. layoutWrap (NO_WRAP | WRAP) applies to HORIZONTAL only. Empty wrappers strip leftover default white fill unless clearFill is false.

Inputs: nodeId (required), layoutMode (required), clearFill, layoutWrap.

set_layout_sizingB

Set hug/fill/fixed sizing. Prefer layoutSizingHorizontal/Vertical or width/height aliases with FIXED | HUG | FILL. For TEXT nodes you can also pass textAutoResize (WIDTH_AND_HEIGHT | HEIGHT | NONE | TRUNCATE) so wrap mode stays aligned with sizing.

Inputs: nodeId (required), width, height, textAutoResize, +4 more.

set_multiple_annotationsC

Batch create/update multiple annotations efficiently.

Inputs: annotations (required).

set_multiple_text_contentsC

Batch update multiple text nodes efficiently.

Inputs: updates (required).

set_node_explicit_variable_modeC

Set an explicit variable mode for a node for a given collection.

Inputs: modeId (required), nodeId (required), collectionId (required).

set_overflow_directionC

Set frame overflow direction for scrolling in prototype (NONE, HORIZONTAL, VERTICAL, BOTH).

Inputs: frameId (required), overflowDirection (required).

set_overlay_settingsB

Configure how a frame/component behaves when it is shown as an OVERLAY: where it's anchored, its scrim background, and whether clicking outside closes it.

Inputs: nodeId (required), overlayBackground, overlayPositionType, overlayBackgroundInteraction.

set_paddingC

Set padding values for an auto-layout frame (top, right, bottom, left).

Inputs: nodeId (required), top, left, right, +1 more.

set_prototype_start_nodeA

Set (or clear, by omitting nodeId) the current page's default prototype start frame — the entry point used by Present.

Inputs: nodeId.

set_reactionsA

Set prototype reactions on a node (replaces existing reactions). Supports every action type including NODE navigation (NAVIGATE/SWAP/OVERLAY/SCROLL_TO/CHANGE_TO), BACK, CLOSE, URL, SET_VARIABLE, SET_VARIABLE_MODE, CONDITIONAL, and UPDATE_MEDIA_RUNTIME. Reactions are validated against a strict schema: transitions of type DISSOLVE/SMART_ANIMATE/SCROLL_ANIMATE must NOT carry direction or matchLayers (only MOVE_IN/MOVE_OUT/PUSH/SLIDE_IN/SLIDE_OUT may), and a NODE destination must be a valid target for its navigation type — NAVIGATE/SWAP need a top-level frame, OVERLAY an overlay-configured frame, SCROLL_TO a node inside a scrollable ancestor, CHANGE_TO a sibling variant in the same component set.

Inputs: nodeId (required), reactions (required).

set_section_propertiesC

Edit properties of an existing SECTION node, e.g. sectionType (SECTION | VIEWPORT) or the raw sectionProperties object.

Inputs: nodeId (required), sectionType, sectionProperties.

set_selectionsA

Set selection to multiple nodes and scroll viewport to show them. Switches to the first node's page; Figma scopes selection to one page, so nodes on other pages are reported in skippedOnOtherPages.

Inputs: nodeIds (required).

set_smart_animate_reactionC

Create or replace a node-to-node prototype reaction using Smart Animate, with optional easing/preset overrides. Smart Animate interpolates matching layers between source and destination automatically; there is no per-property keyframe API.

Inputs: nodeId (required), destinationId (required), match, preset, +9 more.

set_stroke_colorB

Set or clear a node's stroke. Color via hex/strokeHex or r/g/b (0..1 or 0..255), optional strokeWeight, strokeAlign (CENTER|INSIDE|OUTSIDE), dashPattern (e.g. [4,4]), strokeCap (NONE|ROUND|SQUARE), strokeJoin (MITER|BEVEL|ROUND), or styleId to apply a paint style. Pass clear:true to remove strokes.

Inputs: nodeId (required), b, g, r, +10 more.

set_target_frameB

Sets the target frame(s) that the agent is allowed to modify.

Inputs: frameId, frameIds.

set_text_contentC

Set the text content of a single text node.

Inputs: nodeId (required), characters (required).

set_text_styleB

Apply a text style and/or fine-grained typography to a TEXT node. Pass textStyleId to apply an existing style; also supports fontFamily/fontStyle, variationSettings (variable-font axes, e.g. {wght:550}), fontSize, lineHeight (number|'AUTO'|{unit,value}), letterSpacing (number|{unit,value}), textCase, textDecoration, textAlignHorizontal (LEFT|CENTER|RIGHT|JUSTIFIED), textAlignVertical (TOP|CENTER|BOTTOM), textAutoResize (WIDTH_AND_HEIGHT|HEIGHT|NONE|TRUNCATE), layoutSizingHorizontal/Vertical or width/height (FIXED|HUG|FILL), paragraphIndent/Spacing, fillsHex/fills/fillStyleId, boundVariables, textWrapStyle (AUTO|BALANCE|PRETTY), textTruncation (DISABLED|ENDING) and maxLines.

Inputs: nodeId (required), fills, width, height, +22 more.

set_timeline_durationA

Set the duration, in seconds, of the timeline owned by the node's containing top-level frame. Omit timelineId to use the node's first timeline. Lengthen to make room for an animation; do not shorten unless the user asked, since it truncates existing motion.

Inputs: nodeId (required), duration (required), timelineId.

set_transition_reactionA

Create or replace a node-to-node prototype reaction with a typed transition/easing payload. Pass preset for a curated motion preset or transition for an explicit one; direction/matchLayers apply only to MOVE_IN/MOVE_OUT/PUSH/SLIDE_IN/SLIDE_OUT. Set replaceExisting:false to append instead of replacing the node's existing reactions.

Inputs: nodeId (required), destinationId (required), match, preset, +9 more.

set_variable_modeA

Theme switch: set the variable mode for one or many nodes. modeId can be a mode id or exact mode name. Scope via nodeIds, rootNodeId (recurse defaults true, pass recurse:false for the root only), or the whole current page.

Inputs: modeId (required), nodeIds, recurse, rootNodeId, +1 more.

set_variable_valuesC

Update a variable's values by mode.

Inputs: variableId (required), valuesByMode (required), valuesByModeEntries.

set_variant_propertiesC

Rename a component using Figma's variant naming format (Property=Value, ...).

Inputs: properties (required), nodeId, componentId.

set_vector_pathsC

Replace the SVG path data on an existing VECTOR node.

Inputs: nodeId (required), vectorPaths (required).

subscribe_eventsSubscribe to eventsA

Start pushing Figma events (selectionchange, documentchange) to the bridge. They land in the event log read via get_events.

Inputs: events (required).

swap_instance_componentC

Swap an instance to a different library component key.

Inputs: instanceId (required), newComponentKey (required).

undoUndo last actionA

Reverses the most recent auto-captured mutating action (snapshot-based, best-effort). Cannot restore deleted nodes or structural changes. Only works within the current plugin session while target-frame mutating actions were executed through this bridge.

Inputs: no named parameters.

ungroup_nodeB

Ungroup a GROUP node, moving its children up to the group's parent.

Inputs: nodeId (required).

unsubscribe_eventsUnsubscribe from eventsC

Stop pushing selected Figma events to the bridge.

Inputs: events (required).

upsert_reactionC

Replace the first matching reaction (by trigger/action/destination) or append if none match. Supports all action types including NODE navigation. Existing reactions on the node are re-normalized before being written back, so round-tripping does not trip the strict setter schema.

Inputs: nodeId (required), match, reaction.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    C2.9/5.0

    Scored across 188 tools

    Disambiguation2/5

    With 188 tools there are many overlapping purposes: figma_set_solid_fill vs set_fill_color both set solid fills, and text-setting is split across figma_set_text, set_text_content, set_multiple_text_contents, and create_text. Reaction tools (set_reactions, upsert_reaction, set_transition_reaction, set_smart_animate_reaction) and instance-creation tools (create_instance_from_component_key, create_instance_from_set_key, create_instance_from_instance, create_component_instance) also have fuzzy boundaries. Descriptions help, but an agent will frequently misselect.

    Naming Consistency4/5

    The vast majority follow a predictable snake_case verb_noun pattern (create_frame, set_padding, get_node_info, delete_variable). Minor deviations exist: a handful of tools carry a figma_ prefix (figma_set_solid_fill, figma_bridge_status) and a few are not strictly verb_noun (boolean_group, combine_as_variants), but overall it is readable and consistent.

    Tool Count1/5

    188 tools is an extreme mismatch for any single server surface, far beyond the 50+ threshold that indicates an unwieldy set. Even for a rich domain like Figma, this volume forces the agent to navigate a huge catalog and increases the chance of picking the wrong tool.

    Completeness5/5

    The surface covers an enormous breadth: node CRUD, fills/strokes/effects, text, auto-layout and grid, variables/tokens, components and slots, prototyping reactions and motion, styles, comments, exports, and events. Nearly every design workflow has a corresponding operation, with no obvious dead ends.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Empowers AI assistants to control Figma via natural language, enabling creation and modification of designs, components, variables, and exports through a WebSocket bridge.
      2
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI agents to inspect Figma selections, navigate pages, render previews, generate starter code, and submit user-approved canvas edits through a local bridge.
      3 npm
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables a locally run agent or script to read and write a real Figma layer tree through a local bridge — inspecting node structure and variables, exporting screenshots into the model's context, and performing batched create/update/move/rename/delete operations that collapse into a single undo step. All traffic stays on localhost between the bridge and the Figma plugin, with design-token auditing, binding, and creation included.
      MIT