Every generate call hands back a prediction ID rather than a result, so the caller holds state and returns through the prediction tool — quick_generate hides the model lookup and schema fetch but still finishes on that same path. The account tools are the reason to keep this connected rather than calling the API directly: balance and daily cost sit beside the calls that consume them, split by model or key. What you can actually generate is whatever the catalogue holds at the time, which is why the discovery tools come first rather than a hardcoded list.
A front end for the Atlas Cloud API covering model discovery, generation across image, video, 3D and audio, OpenAI-format chat, and the account's balance, usage and cost buckets. The catalogue runs to hundreds of live models across video, image, 3D, LLM and audio.
- Model discovery before you commit: search matches a query against model names, types, providers and tags; the listing filters by Text, Image, Video or Audio, with image-to-3D and text-to-3D arriving under Image; and the detail call fetches a model's OpenAPI schema and returns its input and output shape, pricing and cURL examples (atlas_search_docs, atlas_list_models, atlas_get_model_info).
- Generation handled as a submitted job: each generate call takes a model and parameters and returns a prediction ID immediately, with the output URLs collected later once the status reaches completed or succeeded (atlas_generate_image, atlas_generate_video, atlas_generate_audio, atlas_get_prediction).
- Parameters validated against the model's schema before the request goes out — a missing parameter, a wrong type or one the model does not accept comes back as a precise error and spends nothing.
- One call that collapses the whole sequence: find the model from a keyword, fetch its schema, build the parameters and submit — covering text-to-image, image editing, 3D, image-to-video, lipsync and text-to-speech (atlas_quick_generate).
- A local file turned into a publicly reachable URL, which is the form image-editing and image-to-video models want (atlas_upload_media).
- Speech-to-text over recordings, meetings, interviews, podcasts and voice notes, and OpenAI-format chat completions against the LLM models in the same catalogue (atlas_transcribe_audio, atlas_chat).
- Spend readable in the same session as the spending: available balance with cash, bonus, frozen amount and credit grant status, plus daily UTC usage and cost buckets that group_by splits by model type, model or key (atlas_get_balance, atlas_get_model_usage, atlas_get_model_costs).
Node.js 18 or newer and an Atlas Cloud API key in ATLASCLOUD_API_KEY — the account is free to open. Add it in one line with the CLI agents that support it, or `npx -y atlascloud-mcp` in any client's MCP configuration.
One command plus a key — npx -y atlascloud-mcp, then supply credentials