An MCP server for searching and downloading images using Google Image Search via SerpAPI
This server has basic tool definitions but significant quality gaps. Both tools are explicitly registered with descriptions and parameter schemas visible in main.py. However, descriptions are minimal (under 100 chars), parameter descriptions lack depth and specificity, output schemas are not documented, and error handling provides no recovery guidance. The naming is acceptable (verb_noun format), but the parameter design exposes implementation details rather than abstracting API complexity. No input validation rules, no field constraints, and no structured output documentation. Overall definition quality reflects a prototype-stage implementation, not production-ready.
Download an image to a local directory
Search for images using Google Image Search
Output schemas not documented. search_images_tool returns a list of ImageSearchResult objects (position, thumbnail, source, title, link, original, is_product, size, original_width, original_height) but this structure is not visible in the tool's docstring or return type annotation. LLMs cannot plan downstream tool calls without knowing what fields to expect.
Parameter descriptions are incomplete and lack constraint information. 'limit' has a default but no stated bounds (min/max). 'query' lacks guidance on length, special characters, or expected format. 'image_url', 'output_path', and 'filename' lack validation rules (e.g., does filename need extension? are there path traversal guards?).
Error responses provide no recovery guidance. Both tools catch exceptions and return isError=true with a generic message. For search_images_tool, 'Failed to search for images: ...' gives no hint whether the API key is invalid, the query is malformed, or the service is down. For download_image_tool, a failed download should suggest retry, check URL validity, or verify write permissions, none of which are indicated.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
Tool descriptions are too brief (35 and 30 chars respectively) to provide selection context. 'Search for images using Google Image Search' does not explain when to call this instead of other search tools, what data it returns, or what prerequisites exist (e.g., SERP_API_KEY required). Baseline is 194 chars average for A+ tools.
No result pagination or limits documented. search_images_tool accepts a 'limit' parameter but no guidance on maximum value, performance implications, or whether results may be truncated. If SerpAPI returns thousands of images, the tool should cap results and offer pagination (limit + offset or cursor). Current implementation returns all results up to limit with no indication of total available.
No idempotency guarantees or side-effect declarations. download_image_tool creates files on disk (side effect); if an LLM retries due to timeout, the file may be overwritten. The tool description does not state whether repeated calls with the same URL and filename are safe or destructive, or whether the tool supports a dry-run mode.
No input validation or constraints declared. 'output_path' and 'filename' are free-form strings with no documented restrictions. An LLM could be tricked into passing a path traversal attack (../../etc/passwd) or writing to a sensitive directory. No mention of sanitization.