There is nothing to learn here, which is the point — a model that wants an image mid-conversation should not have to negotiate a dozen parameters to get one. Flux Schnell is the fast variant of the family, so the trade is speed over the maximum quality the larger models reach. The cost lands on your Replicate token, so every generated image is a charge you control.
A single-purpose image generation server. It takes a prompt, runs it through the Flux Schnell model hosted on Replicate, and hands back a URL — no local model, no GPU, nothing to manage beyond an API token.
- One tool taking one required `prompt` string — a text description of the image you want — `generate_image`
- A structured response rather than bare text: a `status` for the request, an `output` carrying the URL of the generated image when it succeeded, and an `error` when it did not
- Input validation before the API call, so a malformed request fails locally instead of costing a round trip
- Named error handling for the cases that actually occur — a missing token, invalid parameters, an API failure, a network problem
A Replicate API token from your Replicate account settings, supplied as `REPLICATE_API_TOKEN` in the `env` block of your client config — never committed to the repository. Node.js: clone into your MCP servers directory, `npm install`, `npm run build`, and point your client at the built entry with node. Generation is billed by Replicate against that token.