<a href="https://modelrunner.ai">ModelRunner</a> is a hosted remote MCP server that lets AI assistants like Claude and Cursor run 100+ AI models — text-to-image, image-to-image, text-to-video, image-to-video, video-to-video, music generation, speech-to-text, and image-to-3D, including Kling, HiDream, Hunyuan Image, Stable Audio, and Rodin. One connection exposes every model as a callable tool: search the catalog, inspect a model's input schema, run inference with run_model, and get results back as hosted URLs directly in the conversation.
Unlike single-vendor MCP servers, ModelRunner aggregates models from multiple providers behind one endpoint and one billing account — pay-per-use credits, no subscription. Authorization is OAuth 2.1: your assistant opens a browser window once; no API key is ever pasted into a config file. Beyond inference, the server can upload local files for use as model inputs and author reusable "wrappers" — simplified, purpose-built interfaces over base models.
Endpoint: https://mcp.modelrunner.run/mcp (Streamable HTTP; registry name ai.modelrunner/mcp)
Setup and full tool reference: <a href="https://modelrunner.ai/mcp">ModelRunner MCP setup guide</a> · <a href="https://github.com/modelrunner/modelrunner-mcp">GitHub</a>