CCAPI Developer Platform

Claude Code MCP: image, video and music generation

Give Claude Code, Cursor, Cline or Windsurf eleven new tools through the Model Context Protocol, using the CCAPI key you already have.

What the Model Context Protocol adds

An AI coding assistant is a language model: it writes code well and cannot render a pixel. Image, video and music generation need entirely different models, which is why producing an asset normally means leaving the editor for a separate tool.

MCP is the open standard that closes that gap. An MCP server publishes a list of tools, the assistant reads that list, and from then on it can call them directly. CCAPI serves MCP from the API gateway itself, so there is no local process to run and no second credential to manage.

Connect in three steps

  • Create an API key at Console then Tokens, and add credit under Top Up.
  • Add one configuration line: for Claude Code, claude mcp add --transport http ccapi https://api.ccapi.ai/mcp with an Authorization Bearer header. Cursor, Cline, Windsurf and Zed take the equivalent JSON block.
  • Restart the editor and ask in plain language, for example generate a hero image of a red bicycle against a whitewashed wall.

Eleven tools across five jobs

  • Images — generate_image, edit_image and upscale_image for prompt-driven generation, editing and 2x super-resolution.
  • Video — generate_video for text-to-video and image-to-video, 4 to 15 second clips with aspect ratio and resolution control.
  • Music — generate_music, generate_lyrics and extend_music for full songs, lyrics alone, or continuing an existing track.
  • Text — chat_completion to reach any text model on the platform for a second opinion or bulk work.
  • Discovery — list_models, get_balance and get_task to resolve model names, check credit and poll long-running jobs. These three are free.

Scoped endpoints keep tool selection reliable

An assistant choosing between four tools is more reliable than one choosing between eleven. Alongside the full endpoint at /mcp, CCAPI publishes /mcp/image, /mcp/video, /mcp/music and /mcp/text, each carrying one capability plus the shared discovery and polling tools.

Connect two if you want two — they compose, and an unknown scope returns a clear error listing the valid ones rather than quietly serving something else.

Asynchronous work is handled for you

Video takes roughly 30 to 180 seconds to render and music 30 to 120, which is longer than most editors will hold a single tool call open. Those tools return a task identifier and the assistant polls until the job finishes, so a pause after submission is the system working rather than a hang.

Generation is billed identically to a direct API call. There is no separate MCP pricing tier and no discount, and model lookup, balance checks and job polling cost nothing.

Frequently asked questions

Do I need to install anything to use the CCAPI MCP server?

No. The MCP server runs as part of the API gateway, so you add a URL and a key. There is no local process, no npx package and no second credential to rotate.

Which AI assistants support this?

Any client that speaks MCP over HTTP transport, which currently includes Claude Code, Cursor, Cline, Roo Code, Windsurf and Zed.

Is MCP more expensive than calling the API directly?

No. A tool call re-enters the same pipeline as an HTTP request, with the same pricing, routing and usage logs. The two are indistinguishable in your dashboard.

Can I use my existing CCAPI API key?

Yes. It is the same key used for REST calls; there is no separate MCP credential to create.

Can I restrict which models the assistant can reach?

Yes. A token can be pinned to a specific model group, and the model list the assistant sees is filtered to what that token can actually call, so it is never offered a model it cannot use.