123 image, video and audio models. From code, or straight from your agent.
Affogato inside Claude, ChatGPT, Cursor and your coding agent.
Add one URL and sign in with your Affogato account. No API key. Every model and workflow, with a quote before each spend and limits you set.
What is MCP? The Model Context Protocol is the open standard that lets an AI assistant use outside tools, so yours can generate with Affogato and spend only what you allow.
Pick your agent.
One path per tool. Chat apps take a click or a pasted URL; coding agents take one command. Every one of them signs in the same way.
Connect Claude.ai
One click, then sign in. No terminal, no key.
- 1
Add Affogato
Add to ClaudeOr by hand: Settings → Connectors → Add custom connector → paste the MCP URL.
https://api.affogato.ai/mcp - 2
Sign in once
The first request opens an Affogato sign-in page. Pick the workspace whose credits the agent may spend and approve it. No API key is created or pasted anywhere.
- 3
Ask for the result
Describe the outcome. The agent picks the model or workflow, quotes the credits, asks before anything large and hands back the file. Things to ask.
Skills and the Claude Code plugin
GitHubAgent Skills that teach any agent the studio workflows, and a Claude Code plugin that installs them together with this MCP server. MIT, at affogato-ai/skills.
/plugin marketplace add affogato-ai/skills
/plugin install affogato@affogatonpx skills add affogato-ai/skillsDescribe the outcome.
Each prompt is a workflow the server delivers in one or two tool calls. Copy, paste, attach the files it mentions.
In Claude or ChatGPT
- “Make a 20-second talking-head video of me reading this script, on the office set.”
- “Put this hoodie on a model in a street scene, three angles.”
- “Swap me into this clip. Here is my photo.”
- “Give me a professional headshot from this selfie, corporate style.”
- “Shoot my product in a kitchen scene for an ad, square and vertical.”
- “Voice this script with a warm female voice, then lipsync it to this video.”
In a coding agent
The agent reads llms.txt and the tool descriptions; these show the shape of a good request.
- “List the models that take an image and return video, cheapest first.”
- “Estimate this generation before running it, and stop if it is over 100 credits.”
- “Generate three variations with nano-banana-pro and save the outputs to ./out.”
- “How many credits does this workspace have left today?”
Sign-in, credits, scopes, models.
OAuth, not keys
The client opens an Affogato sign-in page. You sign in as yourself and choose the workspace the agent may use. No API key is created or pasted anywhere.
Disconnect an agent any time from the Developers page of the app.
The workspace you connect
Generations spend the credits of the workspace you picked when you approved the agent: the same credits as the app and the API.
The workspace needs a paid plan or purchased credits. switch_workspace moves a session to another workspace you belong to.
Three permissions
read List models and read your generations, files and account. generate Spends credits of the workspace you connect. files Add images, video and audio as inputs for generations.
The consent page lists exactly what the agent asked for.
The whole catalog
Every image, video and audio model the API serves, with the same ids, parameters and prices in credits. list_models and get_model are the catalog; estimate_cost quotes before anything is charged.
What the agent can call.
The tools mirror the REST API, studio workflows included. The scope column is the permission each one needs.
| read | list_models | List the models you can call, with filters. |
| read | get_model | One model: parameters, input slots, price in credits. |
| read | estimate_cost | Quote a request in credits without charging. |
| generate | generate | Create a generation. Spends credits. |
| read | get_generation | Read a generation and its outputs. |
| read | wait_for_generation | Block until a generation reaches a terminal status. |
| read | list_generations | Recent generations, including failed and cancelled. |
| generate | cancel_generation | Cancel a queued or processing generation. |
| files | upload_file | Upload a file to use as a model input. |
| read | list_files | List uploaded files. |
| read | get_account | The connected workspace, plan and credit balance. |
| read | switch_workspace | Change which of your workspaces the session spends from. |
| read | enhance_prompt | Rewrite a prompt the way the studios do. |
| read | list_voices | Library voices for text_to_speech. |
| read | list_headshot_styles | Curated headshot styles and the control vocabulary. |
| read | list_talking_head_sets | Curated background sets for talking heads. |
| read | list_product_templates | Scene templates for product ad shots. |
| read | list_characters | Your saved characters. |
| generate | lipsync | Sync a video's mouth to an audio track. |
| generate | text_to_speech | Speech from text with a library or custom voice. |
| generate | talking_head_keyframe | Place a person on a set (talking head, step 1). |
| generate | talking_head_video | Animate a keyframe to a speech track (step 3). |
| generate | headshot | A professional headshot in a curated style. |
| generate | cast_swap | AI Character Swap: replace the people in a clip with your photos. |
| generate | try_on | Garments worn by a person or a described model. |
| generate | product_shot | An ad photo of your product in a scene template. |
Agents spend inside limits you set.
Both limits are per workspace, set by its owners and admins on the Developers page of the app.
Estimate first
estimate_cost quotes a request in credits and never charges; generate reports the charge in its result.
Confirmation above a threshold
A generation above the workspace's confirm threshold is not started until the agent confirms it with you.
Daily cap
Agents stop spending once they reach the daily credit cap set on the Developers page. The cap resets every day.
Failed and cancelled generations are refunded
Credits are charged up front and returned in full when a generation fails or is cancelled.
Bring Affogato to your agent.
Add the URL, approve it once, ask for the result. Agents can read llms.txt; the REST API has an interactive reference.