wtf is chutes-media-mcp?

This is a tool that lets AI coding assistants like Claude Code, Cursor, and Cline generate media files directly inside the project they are working on. The supported media types are images, video, music, and speech. A generated image or audio clip is saved as an actual file in the project folder, so the agent can immediately reference it in code, an HTML page, or anywhere else in the codebase. The tool runs as an MCP server, which is a standard way for AI coding tools to access external capabilities. You register it once in your coding tool's configuration with a Chutes API key, and from that point the agent can discover, describe, and invoke media generation models without manual steps. The workflow follows a describe-then-generate pattern. Because different image and audio models accept completely different settings, the tool first fetches the live input specification for whichever model the agent wants to use. The agent reads that specification and builds the correct request, then sends it. This means the tool never hardcodes parameters for a specific model and stays correct as models are updated. Generated files land in an assets folder inside the current project by default. A small sidecar file is written alongside each asset recording which model was used, what parameters were passed, and a hash of the model's schema at generation time. Unknown or misspelled parameters are rejected before a generation call is made, so mistakes fail early rather than silently. There is also a standalone command-line interface for use in shells and environments that do not support the MCP protocol. Installation is via npm and requires Node.js 20 or later and a Chutes API key. The license is MIT.