CCAPI

What the CCAPI MCP server does

The CCAPI MCP server gives an MCP client access to CCAPI’s catalog of text, image, video, music, and lyrics models. The available model list depends on the CCAPI key, so the server can expose different options for different accounts or key groups. It also reports the key’s used and remaining balance.

Text requests use chat_completion, which sends a conversation to a selected text model and returns a non-streaming reply. The catalog may include providers such as Claude, GPT, Gemini, DeepSeek, GLM, MiniMax, Kimi, and Qwen, but the exact models available to a key must be checked rather than assumed.

Media tools cover several workflows. generate_image creates images from prompts and can use reference images. edit_image modifies an existing image and requires at least one source image. upscale_image enlarges an image by 2x with ccapi-upscale-v1. Music tools can create songs, generate lyrics without audio, or continue an existing song from a timestamp. Video generation accepts text prompts and optional reference images, including first- or last-frame workflows.

How it works

Start with list_models when the model name or capability is uncertain. The response identifies models the current key can call and the MCP tool associated with each one. get_balance can be used before a batch or other costly generation to inspect remaining credit.

Some operations return results directly. Chat completion is non-streaming, and image generation normally returns image URLs after roughly 10–60 seconds. Longer media operations return a task identifier. generate_video, generate_music, and generate_lyrics are asynchronous by design; image generation can also be made asynchronous with async=true. extend_music likewise returns work that must be checked later.

Use get_task with the returned platform and task identifier to follow asynchronous work. Tasks move through queued, in_progress, completed, or failed. The server can poll for a specified wait period, or the client can poll every few seconds until the task settles. A completed music task contains the clip identifier needed by extend_music in its raw payload.

Tools and capabilities

The CCAPI MCP server exposes these operations:

• chat_completion for non-streaming conversations with available text models.
• list_models for key-specific model and capability discovery.
• get_balance for remaining and used credit.
• generate_image for prompt-based or reference-guided image creation.
• edit_image for instruction-based changes to supplied images.
• upscale_image for 2x enlargement of supported images.
• generate_video for asynchronous text-to-video and reference-driven video jobs.
• generate_music for vocal or instrumental songs, with simple and custom lyric modes.
• generate_lyrics for text-only lyric drafting.
• extend_music for continuing a previously generated song.
• get_task for checking asynchronous generation status.

The CCAPI MCP server supports a two-stage music workflow: generate lyrics first for review, then provide approved lyrics to custom music generation. Custom music requests accept lyrics, a title, and style tags; simple mode lets the model write the lyrics.

Limitations and notes

Generation consumes credit. Video is identified as the most expensive capability, and image generation also costs money, so clients should avoid speculative retries and confirm prompts before repeating costly requests. Video jobs commonly take about 30–180 seconds, while music generation usually takes about 30–120 seconds and normally produces two variations.

Image upscaling accepts inputs from 256×256 through 1024×1024 and below 5 MB. Larger images should instead be regenerated with a high-resolution model. edit_image is not the general text-to-image path; use generate_image for pure prompt generation or for models that accept references through generation. The supplied material does not specify installation commands, client configuration, environment-variable names, or a license.