There is nothing to tune here beyond the prompt — no size, seed or step parameters exposed — which makes it trivially easy to wire in and correspondingly inflexible. It bills against your own Replicate account, so image spend shows up where you can see it. If you want control over generation parameters, this is not the server for that; if you want a model to be able to produce an illustration mid-conversation, it is exactly enough.
A small MCP server that calls the Flux Schnell model on Replicate. You give it a prompt, it handles the Replicate API exchange and returns the full API response, which includes the generated image URL and the rest of the metadata.
- `generate_image` — takes a required `prompt` string and returns Replicate's full response, including the image URL
- Detailed prompts pay off: style, scene and detail can all go in the one string
- Error handling around the Replicate exchange, so failures come back as messages rather than silence
A Replicate API token, created at Replicate's API tokens page and passed as `REPLICATE_API_TOKEN` in the server's env block. Node.js v14 or higher. Installation is a clone plus `npm install` and `npm run build`, then pointing the client at `build/index.js` — the package is `flux-schnell-mcp`, 1.0.0 in package.json. Generation takes time; the README notes responses are not instant. Documentation is in Chinese.