Image generation MCP server for every image model. One API key, one balance, every major model.
Four tools with complete schemas, clear descriptions, and proper error handling. Naming follows verb_noun convention (generate_image, edit_image, list_models, get_balance). All parameters have types and descriptions. Schemas use Zod with regex patterns and length constraints. Error responses include actionable guidance (e.g., 'top up at https://pixfaro.com/topup'). Low balance warnings are user-friendly. However, descriptions are concise but could better explain WHEN to use each tool vs alternatives, and output schemas are not formally documented. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels (WRITE vs READ_ONLY).
Edit a previously generated image with a natural-language instruction (e.g. "make the sky darker", "remove the text"). Pass the img_… id returned by generate_image.
Generate an image from a text prompt. Returns a hosted image URL (valid long-term) plus cost and balance. Pick a model with list_models; nano-banana-2 is the general-purpose default.
Show the current prepaid balance in USD.
List available image models with pricing, latency, and what each is best for.
Output schemas not formally documented. Tool descriptions state what is returned (e.g., 'hosted image URL plus cost and balance') but do not specify the exact JSON structure or field types. LLMs cannot reliably extract fields from unstructured text responses.
Tool annotations missing. generate_image and edit_image are destructive (WRITE risk) but lack destructiveHint annotation. list_models and get_balance are read-only but lack readOnlyHint. Annotations help agents reason about side effects and retry safety.
No idempotency guidance. generate_image and edit_image modify state (charge balance, create images). Descriptions do not clarify whether repeated calls with identical parameters are safe or will incur duplicate charges. Agents need this to decide retry strategy.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
Error handling lacks recovery paths for common failures. 'insufficient_balance' error includes a topup URL, but other errors (invalid model, malformed image ID, API timeouts) do not suggest next steps. LLMs need actionable guidance for each error class.
Tool descriptions lack context on when to call each tool. 'list_models' description does not explain that agents should call it first to discover available models before calling generate_image. 'get_balance' does not clarify when to check balance (before expensive operations, after failures, etc.).