MCP server
Generate from Claude, ChatGPT, Cursor or any other MCP client, on your own account.
Gensprite is also a Model Context Protocol server. Add it to an AI assistant and the assistant can draw sprites, turn characters around, animate them and make props, materials and plain images for you — the same jobs as the workshop, charged to the same credits, saved to the same history.
https://api.gensprite.ai/mcpConnecting
Add the URL as a remote (HTTP) MCP server or custom connector in your client. The first time it connects, the client opens a Gensprite page in your browser: sign in, check what it is asking for, and allow it. Nothing to copy, and no key to keep safe.
-
Claude (web and desktop): add a custom connector with the URL above.
-
Claude Code:
claude mcp add --transport http gensprite https://api.gensprite.ai/mcp -
Cursor: add it to
mcp.json:{ "mcpServers": { "gensprite": { "url": "https://api.gensprite.ai/mcp" } } } -
ChatGPT and other clients: wherever they let you add an MCP server by URL.
With an API key instead
Clients that cannot open a browser, and scripts, can send an API key rather than signing in:
Authorization: Bearer gsk_...For example claude mcp add --transport http gensprite https://api.gensprite.ai/mcp --header "Authorization: Bearer gsk_...",
or "headers": { "Authorization": "Bearer gsk_..." } next to url in Cursor.
What it can do
When you allow a client you choose between two permissions:
| Permission | Lets the client |
|---|---|
| Read | See your balance, models and prices, projects, characters and past generations. |
| Generate | Start generations, which spend your credits, and create projects, characters and uploads. |
A client that was only allowed to read asks again the first time it tries to spend. An API key always has both.
The tools it gets:
generate_sprite,rotate_character(1, 4 or 8 views),animate_character,generate_prop,generate_material— each takesquote_onlyto show the price first.generate_imageis raw mode: the prompt goes to GPT Image 2.5, Nano Banana 2.1 or Seedream 5.0 Pro exactly as written, with no sprite framing.background: "transparent"works with all three. It takesquote_onlytoo.generate_musicandgenerate_soundmake music and sound effects, looping ones withloop: true, and return WAV, OGG and MP3 of the same audio. Both takequote_only.generate_3d_characterturns a picture (a sprite's front view works best) or a description into a rigged 3D character, andanimate_3d_characteradds a motion described in words to it. See 3D characters and motion. Both are for paying accounts;quote_onlyworks for everyone.get_jobwaits for a generation to finish and returns its files.list_jobsandcancel_job(which refunds) manage the rest.get_account,list_models,list_projects,create_project,list_characters,create_character,attach_job,create_upload.search_docsandread_docsread this documentation. Before writing a prompt, the assistant is told to read the guide for that kind of generation, such as Making a sprite or Animating. When a result comes out wrong, it is told to search here before trying again.
A character usually goes sprite, then a four-way turnaround of that sprite, then animations of the four views — ask for it in those words and the assistant chains the tools itself.
Limits
Images given to a tool must already be Gensprite files: the output of an
earlier generation, or an image you uploaded with create_upload, which needs
a client that can make HTTP requests (Claude Code and Cursor can). Attaching a
picture in a chat and asking to animate it does not work yet.
File URLs in results are time-limited, like everywhere else in the API. Download what you want to keep.