MCP App Store
by KeenEthics
Toonkit icon

Toonkit MCP Server for ChatGPT

By Innerverz

Overview

Create images and videos with Toonkit Playground, or organize text, media and generation recipes in Toonkit Canvases. Browse saved assets, inspect supported models, estimate costs and track generation results. Canvas tools also support voice generation and 3D Reference scene editing and export. Requires a Toonkit account and authorized MCP connection. AI generation and retries spend Toonkit credits; estimates and Canvas editing do not. Queued Canvas edits and 3D rendering require a compatible connected Toonkit browser tab to apply or finish.

How to connect

  1. ChatGPT: Open the Toonkit listing in the ChatGPT app directory and choose Connect. ChatGPT listing Toonkit exposes 49 tools here.

Tools

toonkit_canvas_add_generation_node

ChatGPT
Add a generation recipe node WITHOUT running it. Nothing is charged and nothing is generated — use toonkit_canvas_generate_* to actually run. Does not spend credits.

toonkit_canvas_add_media_node

ChatGPT
Add an existing media file as a Canvas node. Requires read access to that media. For a local file, first create the media: toonkit_create_upload -> HTTP PUT to the returned uploadUrl -> toonkit_confirm_upload -> use the resulting mediaId here. Files cannot be replaced in place — add a new node and reconnect. Does not spend credits.

toonkit_canvas_add_text_node

ChatGPT
Add a text node to an owned solo Canvas. Does not spend credits.

toonkit_canvas_connect

ChatGPT
Connect compatible nodes without creating a cycle. Does not spend credits. Connecting a media or text node into a generation node's prompt inserts a reference chip (@Image 1, @Video 1, ...) at the head of that prompt — write the prompt FIRST, connect after, because a later prompt patch drops the chips while keeping the edges.

toonkit_canvas_delete_nodes

ChatGPT
Delete selected nodes and attached edges. Generation jobs remain recorded. Does not spend credits.

toonkit_canvas_disconnect

ChatGPT
Remove an edge. Already absent edges are a no-op. Does not spend credits.

toonkit_canvas_estimate_upscale_credits

ChatGPT
Quote Topaz upscaling for a completed image or video node in an owned solo Canvas without spending credits. Pass exactly one of factor (2 or 4) and targetResolution (2K or 4K).

toonkit_canvas_generate_image

ChatGPT
Generate ONE Canvas image, spending up to maxCredits. Returns producer/result IDs. Reuse the key on retry. Style models such as toonxl require styleId from the catalog; passing one ref switches them to image-to-image. toonxl takes a prompt (T2I) XOR exactly ONE reference image (I2I) — passing both, or neither, is rejected. Only I2I ignores the prompt (the style checkpoint carries its own); in T2I the prompt is the ONLY thing that says what to draw, so ask the user for it instead of inventing one.

toonkit_canvas_generate_music

ChatGPT
Generate Suno music on an owned solo Canvas: one producer and two indexed audio nodes. Quote kind=music first. Reuse idempotencyKey on retries. Partial success preserves every actual track.

toonkit_canvas_generate_video

ChatGPT
Generate ONE Canvas video, spending up to maxCredits. Returns producer/result IDs. Reuse the key on retry. Reference kinds are separate inputs: refImageNodeIds, refVideoNodeIds, refAudioNodeIds. Each array is ordered — entries are named @Image 1, @Image 2, ... and @Video 1, ... by position, and mentioning those names in the prompt improves reference adherence for R2V models. Read the video catalog first — model id, resolution and duration must come from specByMode.

toonkit_canvas_generate_voice

ChatGPT
Generate ONE Canvas voice clip from text, spending up to maxCredits. Returns producer/result IDs. Reuse the key on retry. Pick voiceId with toonkit_list_voices. Reference nodes are not accepted.

toonkit_canvas_group_nodes

ChatGPT
After completing one user request, group its intermediate nodes, producers and FINAL output nodes. Collect their IDs from creation/generation/export receipts and pass all of them in workflow order. Requires existing ungrouped non-group nodes. Arranges like the group's Arrange horizontally: one column per connection step, parallel branches stacked, input order for ties, and places the group clear of other nodes. Returns the group nodeId. No credits. Poll get_canvas_mutation for completion.

toonkit_canvas_list_models

ChatGPT
List real Canvas models, supported inputs and credit prices for a kind (image, video or voice). Read this before generating — model ids, aspect ratios, resolutions and durations all come from here.

toonkit_canvas_reference3d_catalog

ChatGPT
For 3D Reference work, use browser tools to open the Canvas and the target 3D Reference node's editor. Keep the target visible while editing and verify the applied result in the browser; a PENDING receipt is not completion. For a new node, open the Canvas first, create the node, wait for it to appear, then open its editor. For export, operate the 3D Reference editor's Export control in the browser and verify the output node appears. If browser tools or access are unavailable, explain the blocker rather than claiming the work is complete. List the 3D Reference editing vocabulary — operations with their domain and fields, object kinds, per-kind joints, motion presets, initialize templates and limits. Read this before toonkit_canvas_reference3d_edit: a batch must use a single domain, and rejected commands are never guessed at. Does not spend credits.

toonkit_canvas_reference3d_create

ChatGPT
For 3D Reference work, use browser tools to open the Canvas and the target 3D Reference node's editor. Keep the target visible while editing and verify the applied result in the browser; a PENDING receipt is not completion. For a new node, open the Canvas first, create the node, wait for it to appear, then open its editor. For export, operate the 3D Reference editor's Export control in the browser and verify the output node appears. If browser tools or access are unavailable, explain the blocker rather than claiming the work is complete. Admit creation of one 3D Reference node — the canvas gains a node whose scene is initialized from the template (empty or human_camera) once a connected Canvas tab applies the intent; call toonkit_get_canvas_mutation to follow it. Reusing an idempotency key returns the original receipt instead of creating twice. Does not spend credits.

toonkit_canvas_reference3d_edit

ChatGPT
For 3D Reference work, use browser tools to open the Canvas and the target 3D Reference node's editor. Keep the target visible while editing and verify the applied result in the browser; a PENDING receipt is not completion. For a new node, open the Canvas first, create the node, wait for it to appear, then open its editor. For export, operate the 3D Reference editor's Export control in the browser and verify the output node appears. If browser tools or access are unavailable, explain the blocker rather than claiming the work is complete. Admit a batch of scene commands for one 3D Reference node — read the vocabulary with toonkit_canvas_reference3d_catalog first, and echo the get_scene revision back as expectedRevision so a stale base is rejected instead of guessed. Batches must use a single domain; scene.initialize needs none. A connected Canvas tab automatically applies the intent; call toonkit_get_canvas_mutation to follow it. Read get_scene(view=logical) before editing; if the revision becomes stale, refetch and reassess before resubmitting. Frame indices are 0 through durationSeconds*fps-1; same-batch aliases use clientRef, not objectId. Does not spend credits.

toonkit_canvas_reference3d_export

ChatGPT
For 3D Reference work, use browser tools to open the Canvas and the target 3D Reference node's editor. Keep the target visible while editing and verify the applied result in the browser; a PENDING receipt is not completion. For a new node, open the Canvas first, create the node, wait for it to appear, then open its editor. For export, operate the 3D Reference editor's Export control in the browser and verify the output node appears. If browser tools or access are unavailable, explain the blocker rather than claiming the work is complete. Queue an image or video export of a 3D Reference node, including while no browser is open. A compatible Canvas tab automatically opens the target 3D editor and renders through its browser runtime. kind=video renders the duration; kind=image captures the editor frame and requires cameraObjectId. Echo expectedRevision from get_scene (view=logical includes pending creation and edits). Rendering uses the target scene when execution starts, including current edits if its editor is open. Follow completion with toonkit_get_canvas_mutation. Rendering does not spend credits.

toonkit_canvas_reference3d_get_scene

ChatGPT
For 3D Reference work, use browser tools to open the Canvas and the target 3D Reference node's editor. Keep the target visible while editing and verify the applied result in the browser; a PENDING receipt is not completion. For a new node, open the Canvas first, create the node, wait for it to appear, then open its editor. For export, operate the 3D Reference editor's Export control in the browser and verify the output node appears. If browser tools or access are unavailable, explain the blocker rather than claiming the work is complete. Read one 3D Reference node's scene — safe object summaries, materialization state (missing node, missing scene, or pending ADD), a signed revision to echo back as expectedRevision on edit, and compatibility flags when stored values needed normalization, the scene holds unsupported objects, or private fields were redacted from the safe view. view=saved (default) is the durable graph; view=logical also folds still-pending 3D intents and reports the first broken dependency as a conflict instead of guessing success. Omit nodeId to read whichever node the user has open and armed for AI edits. Private media URLs and NPZ requests are never exposed. Does not spend credits.

toonkit_canvas_seedance_draft_finalize

ChatGPT
Create a separate Canvas result at 1080p from its Seedance Draft original, spending up to maxCredits. Reuse idempotencyKey on retry. Keeps the original Draft and creates new nodes.

toonkit_canvas_seedance_draft_quote

ChatGPT
Check a Canvas Seedance Draft original and quote its separate 1080p final generation. No credits charged.

toonkit_canvas_update_generation_node

ChatGPT
Patch allowed recipe fields on a generation node. Editing detaches the node from the MCP job that created it, because it is no longer that run's copy. Patching prompt or reference fields can drop reference chips (@Image 1, ...) while keeping the edges, and invalidates a pending generation on that node. Does not spend credits.

toonkit_canvas_update_text_node

ChatGPT
Update an existing text node. Does not spend credits.

toonkit_canvas_upscale_media

ChatGPT
Upscale ONE completed image or video node in an owned solo Canvas with Topaz, spending up to maxCredits. Pass exactly one of factor (2 or 4) and targetResolution (2K or 4K). Preserves the source and creates a connected result node. Returns jobId and resultNodeId. Poll toonkit_get_generation for completion; reuse the same idempotencyKey on retry. Canvas node application requires a compatible ToonKit browser tab.

toonkit_confirm_upload

ChatGPT
Confirm an upload after the PUT succeeded. Returns the mediaId to pass as a generation reference. Reuse the same idempotency key if the response is lost — the stored mediaId comes back. Does not spend credits.

toonkit_create_canvas

ChatGPT
Create an empty Canvas. Requires an idempotency key. Never spends credits.

toonkit_create_upload

ChatGPT
Start an image upload for use as a generation reference. Returns a presigned PUT url and the exact headers you must send — the signature covers Content-Type and Content-Length, so a mismatch is a 403. Then call toonkit_confirm_upload. Images only. Reuse the key on retry. Does not spend credits.

toonkit_delete_canvas

ChatGPT
Soft-delete the entire owned solo Canvas project. Does not cancel paid generation or delete files.

toonkit_estimate_credits

ChatGPT
Quote the credits a generation would cost, using the same model, reference and pricing rules as the real launch. Nothing is charged and no job is created. Pass the same reference nodes you will launch with — reference count and video length are billing inputs, so omitting them changes the quote.

toonkit_estimate_upscale_credits

ChatGPT
Quote Topaz upscaling for an existing owned image or video without spending credits. The connection also needs the source surface's read scope (e.g. assets:read for library media). Pass exactly one of factor (2 or 4) and targetResolution (2K or 4K). Use the quote as maxCredits.

toonkit_generate_image

ChatGPT
Generate ONE image without a Canvas, spending up to maxCredits. Reuse the key on retry. Reference images must already exist in Toonkit — this tool does not accept file bytes. Every requested reference must resolve or the request is rejected, never silently dropped. Style models such as toonxl require styleId from the catalog; passing one ref switches them to image-to-image. toonxl takes a prompt (T2I) XOR exactly ONE reference image (I2I) — passing both, or neither, is rejected. Only I2I ignores the prompt (the style checkpoint carries its own); in T2I the prompt is the ONLY thing that says what to draw, so ask the user for it instead of inventing one.

toonkit_generate_music

ChatGPT
Generate Suno music in Playground; one request expects two tracks. Read models kind=music and quote first. Reuse idempotencyKey after an unknown result. Poll toonkit_get_generation for every actual output.

toonkit_generate_video

ChatGPT
Generate ONE video without a Canvas, spending up to maxCredits. Reuse the key on retry. Reference media must already exist in Toonkit. Reference video length is measured server-side, so do not pass it — it is a billing input for reference-to-video models. Reference arrays are ordered: entries are named @Image 1, @Video 1, @Audio 1, ... by position — mention them in the prompt for R2V models.

toonkit_get_canvas

ChatGPT
Read the logical Canvas, including pending edits and generated results. Does not spend credits.

toonkit_get_canvas_mutation

ChatGPT
Read what happened to one queued Canvas edit. Edits are applied by the browser, so PENDING means no tab has processed it yet. Does not spend credits.

toonkit_get_credit_balance

ChatGPT
Read the credit balance plus BOTH MCP daily limits (account-wide and this connection) with today's usage. remainingToday is the smallest remaining allowance across wallet balance and any configured daily limits; an unset limit means no cap on that axis. Zero when MCP is disabled. A null connection limit uses only the account budget. The daily limits are separate from the wallet — web usage does not consume them.

toonkit_get_generation

ChatGPT
Read one generation job: status, billing, result node, the settings it was admitted with (inputSnapshot — change those and call a generate tool again), and a short-lived URL for the produced media. Call this again when the URL expires — do not cache it. Does not spend credits.

toonkit_get_generation_guide

ChatGPT
Read the ToonKit generation guide once before generating: non-negotiable billing and idempotency rules, surface differences, model-selection rules (incl. toonxl styles and Seedance 2.5 Anime), and Canvas reference-chip mention syntax. Model ids and prices themselves come from the list_models tools. Does not spend credits.

toonkit_get_media

ChatGPT
Get a short-lived URL for one media item you own. Call again when it expires — do not cache it. Does not spend credits.

toonkit_list_assets

ChatGPT
List saved assets (Global Assets), newest first. Use mediaId from a row as a reference input for generation tools. Folders and text rows have no mediaId. Does not spend credits.

toonkit_list_canvases

ChatGPT
List your own Canvases, newest first. Collaborative Canvases are marked and reject MCP writes.

toonkit_list_generations

ChatGPT
List your MCP generation jobs, newest first, with status and billing. Does not spend credits.

toonkit_list_voices

ChatGPT
List voices available for toonkit_canvas_generate_voice. Filter by label. Does not spend credits.

toonkit_playground_estimate_credits

ChatGPT
Quote the credits a Playground generation would cost, using the same model and pricing rules as the real launch. Nothing is charged. Pass the same reference media you will launch with — reference count and video length are billing inputs.

toonkit_playground_list_models

ChatGPT
List Playground models and their limits for a kind (image or video). Read this before generating — Playground models are a DIFFERENT set from Canvas models, so Canvas model ids may not work here.

toonkit_rename_canvas

ChatGPT
Rename an owned solo Canvas. Does not spend credits.

toonkit_retry_generation

ChatGPT
Retry one FAILED Playground generation with the SAME settings, spending up to maxCredits. This charges again — a retry is a new generation, not a free redo. Succeeded generations cannot be retried: to change settings, read inputSnapshot from toonkit_get_generation and call the generate tool again. Reuse the key on retry.

toonkit_seedance_draft_finalize

ChatGPT
Create a separate 1080p final from a Playground Seedance Draft original, spending up to maxCredits. Reuse idempotencyKey on retry. Keeps the original Draft. Read quote first.

toonkit_seedance_draft_quote

ChatGPT
Check a Playground Seedance Draft original and quote its separate 1080p final generation. No credits charged. Source expires seven days after provider creation.

toonkit_upscale_media

ChatGPT
Upscale ONE existing owned image or video with Topaz, spending up to maxCredits. The connection also needs the source surface's read scope (e.g. assets:read for library media). Pass exactly one of factor (2 or 4) and targetResolution (2K or 4K). Returns jobId; poll toonkit_get_generation for the result. Reuse the same idempotencyKey on retry. Creates a new Playground result and preserves the source.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/toonkit"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/toonkit)