AutoSprite

MCP Integration

Connect AutoSprite to Claude Code, Codex, Cursor, Windsurf, and more. Generate sprite sheets from your editor with the Model Context Protocol.

The AutoSprite MCP (Model Context Protocol) server connects your AI coding assistant directly to AutoSprite. Describe a character in chat and your assistant can create it, animate it, and export a game-ready sprite sheet — without leaving your editor.

ℹ️

MCP access is included with a subscription plan. Create your key on the API Keys page — the same key works for both MCP and the REST API.

Setup

Get an API key

Create a key on the API Keys page. Your key starts with vspk_.

Add the configuration

Add AutoSprite to your editor's MCP config. For Claude Code (~/.claude.json):

{
  "mcpServers": {
    "autosprite": {
      "type": "http",
      "url": "https://www.autosprite.io/api/mcp",
      "headers": {
        "Authorization": "Bearer vspk_your_key_here"
      }
    }
  }
}

Other editors use the same URL with a slightly different config shape — the MCP setup page has copy-paste configs for every editor, with your key filled in automatically.

Restart your editor

Reload or restart your editor to connect. You're ready to generate.

Supported editors

Claude Code, Cursor, Windsurf, Cline, Gemini CLI, Codex, and Antigravity — anything that speaks the Model Context Protocol over HTTP.

Available tools

Once connected, your assistant gains these tools:

Account

  • get_account — Get your credit balance. Call at most once per session; the balance only changes when you run a generation tool, which already reflects the outcome. Don't poll.

Characters

  • list_characters — List your characters with pagination and search
  • get_character — Get details for a character, including its spritesheets
  • create_character — Create a character from a text prompt (pixelArt: true for a crisp 16-bit pixel-art sprite, 3 credits)
  • upload_character — Create a character from your own image (URL, base64, or uploadKey) — free
  • request_upload_url — Get a presigned URL to upload a local image, then pass the uploadKey to upload_character
  • generate_portrait — Generate a close-up portrait of a character's face and shoulders
  • generate_pose — Generate a new pose for a character from a prompt
  • pad_character — Add padding around a character's base image to resize it within the frame

3D Models

  • list_3d_animations — List every generatable animation clip with its credit cost (no character required — use for discovery)
  • generate_character_3d_model — Build a rigged 3D model (GLB) from a character's image, including an idle animation (50 credits). Async — returns a jobId; poll get_job_status, then get_character_3d_model for the GLB
  • generate_character_3d_animations — Unlock more animation clips (walk, sprint, sword, jump, etc.) on a rigged model — 5 credits per new animation. Async — returns a jobId
  • get_character_3d_model — Get the model status (not_started / generating / ready), the in-flight activeJobId (to resume polling), the rigged GLB download URL, unlocked animations, and the catalog of generatable animations

Spritesheets

  • generate_spritesheet — Generate animations and queue a sprite sheet export. videoTier picks the video model + price (turbo 5 on our self-hosted MiniMax-H3 engine — recommended; pro 10, ultra 10/sec, max 35); durationSec sets clip length. turboMode (turbo only) is "reference" by default — animates straight from the character art — or "first_frame" to paint a seed frame first (auto-selected when a pose, last-frame pose or first_frame_quality is given). Legacy: videoTier: "pro" with durationSec: 2 runs the old turbo engine (Seedance 1.0) — less consistent than turbo, only for matching animations made before September 2026. Per-animation, set first_frame_quality: "pro" for a sharper pro first frame (+3 credits). Besides idle/walk/run/jump/attack/custom, kind takes any preset — death, victory, punch, dance, sleep, … (same price, no prompt needed)
  • list_animations — List the preset kinds and every weapon's attack moves, with a looping preview of each move (no character required — use for discovery)
  • generate_attack_moves — Side-view weapon attacks, the same moves as the app's attack wizard: pick a weapon (sword, spear, axe, hammer, greatsword, staff, dagger, bow) and 1–4 of its moves — 5 credits per move, always Turbo. The moves of one call share one guard frame, so the weapon looks the same in all of them. Each move's cell is twice frameSize (the character keeps its idle size). Async — returns a jobId per move
  • generate_isometric_pack — Animate an isometric / top-down character: one render produces all eight directions as eight sprite sheets plus a zip, 15 credits per action (idle, walk, run, jump, attack with an optional weapon, a preset such as death, or a custom action). Async — returns a jobId; poll get_job_status
  • regenerate_spritesheet — Regenerate sheets at different sizes from existing videos (free)
  • regenerate_single_spritesheet — Regenerate one sheet from one existing video at a new size (free)
  • get_spritesheet — Get sprite sheet details and download URLs
  • list_spritesheets — List all sprite sheets for a character

Jobs

  • get_job_status — Check the status of a generation job
  • list_jobs — List recent jobs with filtering

Assets

  • list_assets — List static assets (props, objects)
  • get_asset — Get asset details
  • get_asset_spritesheet — Get the sprite sheet for an asset
  • generate_asset_preview — Generate preview images (turbo = 4 fast drafts, ultra = 1 polished, pixelArt: true = 1 crisp pixel-art sprite; 1 credit)
  • create_asset — Create a static asset from a URL or uploadKey
  • animate_asset — Generate an animation video for an asset. videoTier picks the model + price (turbo 5, pro 10, ultra 10/sec, max 35); durationSec sets clip length
  • generate_asset_spritesheet — Generate a sprite sheet from an asset animation
  • generate_asset_3d_model — Generate a 3D model from an asset image
  • remove_asset_background — Remove the background from an asset's image
  • get_asset_job_status — Check the status of an asset generation job

Example prompts

Create a character with animations

"Create a pixel art wizard with purple robes, then generate idle,
walk, and cast-spell animations. Export as a 64x64 sprite sheet."

Batch-generate across your library

"List all my characters and generate walk animations for any
that don't have them yet."

Upload a local image

"Upload my character image at ./assets/knight.png and generate
idle and walk animations."

The assistant will call request_upload_url, upload the file, call upload_character with the uploadKey, then generate the animations.

Side-scroller combat set

"Give my knight death and victory animations, plus the greatsword
Heavy Cut and Titan Cleave attacks."

The assistant calls generate_spritesheet with kinds death and victory, then generate_attack_moves with weapon: "greatsword" and both moves in one call, so they share the same guard frame.

Isometric / top-down RPG

"Make isometric walk and run packs for my knight, then give him
a sword attack."

The assistant calls generate_isometric_pack once per action. Each pack renders all eight directions — up, northeast, right, southeast, down, southwest, left, northwest — for 15 credits, and get_job_status returns the eight sprite sheets plus a zip with every direction. Packs are recommended because every direction comes from the same render, so the character stays consistent. If you only want one facing ("just a down-facing idle"), say so — the assistant generates that single direction with generate_spritesheet (e.g. kind iso_idle_down), billed at the video tier price.

Prefer plain HTTPS?

Building a backend pipeline instead of working in an editor? The REST API exposes the same capabilities over HTTPS.