Generation servers usually make you find the spend afterwards, in a dashboard, once it has happened. Here the credit cost comes back with the result and the balance is one call away, which turns "can I afford to iterate on this" into something the agent can reason about mid-task. The sandbox key prefix is the other detail worth using: you can exercise the whole flow, including the tools that would cost real money, before spending anything. The default surface is deliberately 10 tools so the context stays light; the presentation and video-composition tools sit behind feature gates.
A hosted MCP server for Clipia, an AI creation platform. One endpoint fronts 60+ generation models, Western and Chinese, so an agent can make an image, render a video, produce a voiceover or build a deck without an SDK or a separate subscription per provider. Every response carries the exact credit cost of what it just did.
- Images from a prompt, optionally with reference images for editing or image-to-image, usually returned inline with a preview — `generate_image`
- Video from a prompt or from a start image, returning a request ID and cost immediately since renders take 1–10 min, then polled to completion — `generate_video`, `wait_generation`, `get_generation`
- Speech from text in a chosen voice and language as MP3, and background music from a description of mood, genre and tempo — `generate_audio`, `generate_music`
- The model catalogue with capabilities and pricing in credits, and per-model detail including input schema and context limits — `list_models`, `get_model`
- 3500+ curated prompt templates searchable in Russian or English, each with a ready-to-use prompt and a recommended model — `search_templates`
- Your credit balance and the current key's 30-day usage — `get_balance`
- Feature-gated additions on top of the 10 core tools: chat with a text model, a brief turned into per-scene video prompts, 2–20 finished scenes stitched into a final MP4 with voiceover and subtitles, and editable PPTX decks generated and edited — `chat`, `generate_scenario`, `compose_video`, `generate_presentation`, `edit_presentation`
A Clipia account with credits. IDEs and CLIs authenticate with an API key from the Developer Console against `https://mcp.clipia.ai/mcp` over Streamable HTTP — `Authorization: Bearer clipia_...` is the recommended header scheme. claude.ai and ChatGPT connect over OAuth instead, with no key to manage. Keys prefixed `clipia_test_` run in a sandbox that returns instant mock results and spends no credits, which is the right way to wire up an integration. New accounts get a small pack of welcome credits; after that a subscription is required.
One command plus a key — claude mcp add --transport http clipia https://mcp.clipia.ai/mcp \ --header "Authorization: Bearer <CLIPIA_API_KEY>", then supply credentials