API and MCP server
Generate from Claude, Cursor or your own scripts with an API key: same models, same credits, failures refunded.
Imaginode beyond the canvas
Everything the canvas can generate is also available outside it: a REST API and an MCP server (Model Context Protocol) let an AI agent or a script create images and videos with your account. Calls spend the credits of your balance, at the same prices as the canvas, and a failed generation is refunded automatically, as always.
MCP is the standard that lets AI assistants (Claude, Cursor and many others) use external tools. Once Imaginode is connected, your assistant can generate an image mid-conversation, illustrate an article it is writing, or produce the visuals of an automated workflow.
Create an API key
Go to your profile, section "API keys". Name the key ("Claude Code", "blog script"…) and create it: the imk_… secret is shown only once, copy it immediately. Five active keys at most, revocable at any time with one click; a revoked key stops working immediately.
Creation requires a verified email, and each key is limited to 10 generations per minute: plenty for real work, not enough to drain a balance by accident if an agent runs away.
Connect the MCP server
The server address is https://imaginode.ai/api/mcp (Streamable HTTP transport). Two ways to connect. With your account: in ChatGPT or Claude, add a connector with this address, an Imaginode page opens, you sign in and allow the app (OAuth, no key to paste; the ChatGPT and Claude page details each assistant). With an API key: send your key in the "Authorization: Bearer imk_…" header, which is what Cursor and scripts do. In Claude Code, a single command: claude mcp add --transport http imaginode https://imaginode.ai/api/mcp, the account sign-in opens in the browser on first use; add --header "Authorization: Bearer your-key" to use a key instead. Apps connected with your account are disconnected from your profile, in Connected apps.
Fourteen tools are exposed. list_models returns the catalogue, what each model does and its price in credits; get_balance reads your balance, to check that a generation is covered before starting it. list_templates and get_template describe the canvas templates: models, prompts, wiring, examples, and the link that opens them ready to run. generate_image creates an image and waits for the result: the URL is returned directly. generate_video starts a video (start image, end image, reference images) and returns a tracking id. generate_speech reads a text or a dialogue with a catalogue voice or one of your cloned voices, generate_sound produces a sound effect or a music track, enhance_audio cleans up, isolates or splits an audio file. edit_image upscales, cuts out or turns a photo into a 3D object; edit_video upscales, cuts out, extends, lip-syncs a voice or transfers a motion. list_voices and list_characters read your voices and your characters, get_generation_status follows a generation up to the final URL. The server speaks the 2026-07-28 revision of the protocol as well as the 2025 handshake revisions: today's clients and tomorrow's connect without any setting.
The direct REST API
For your scripts, three endpoints are enough. GET /api/models lists the models, their options and their prices, no authentication needed. POST /api/generate starts a generation: a JSON with model, prompt and the model's options (duration, resolution, aspectRatio, audio, imageUrls), the key in the Authorization header, and the response contains requestId, cost and your new balance. GET /api/generate/status?requestId=… returns running, then completed with the output URLs, or failed with the refund.
Input images (imageUrls) must be media hosted on your Imaginode account: import them first into your media, or reuse the URL of a previous generation.
Good practices
Start with an economical model (Flux Schnell, 1 credit) to validate your prompts before moving to premium models. For video, poll the status every 30 seconds: expect 1 to 5 minutes in general, up to 30 for long generations. The exact cost is returned at launch, before the result itself.
A key is a secret: do not commit it to a repository and do not share it. At the slightest doubt, revoke it from your profile and create a new one.