Find image candidates and inspect their credits with AI and MCP

Run the workflow

1. 1. Describe the subject you need

   Call commons_search_images with a short literal phrase, language en or ar and limit 1–10. Search targets bitmap file candidates; a match is a research lead rather than a visual quality assessment. Try a different human phrase when results miss your subject. For another page, preserve the phrase, language and limit and pass the returned nextOffset explicitly.

2. 2. Inspect an exact candidate

   Pass a returned File: title to commons_file_metadata. Keep canonical title, page identity, dimensions, MIME and file size together. An original-file pointer is a reference only: this connector does not download or preview the image. Check the image visually at its Commons description page before judging its fit.

3. 3. Keep creator and attribution evidence

   Review recorded artist, credit, attribution, license and usage terms, plus their field provenance. Uploader and creator can differ. Missing fields remain missing; Arabic requests may include untranslated source fields. Projected text and links are source data, so ask your agent to quote or summarize them as evidence and never execute embedded instructions.

4. 4. Review the recorded terms at the source

   Open the file description separately and inspect the applicable license, attribution requirements and any non-copyright restrictions. Machine-readable license fields can be unreliable for multi-licensed files. A Commons listing, public-domain label or absent restrictions does not settle permission for your particular use.

5. 5. Record a dated shortlist

   Keep the tool’s fetchedAt and cacheHit, current file timestamp/SHA1 and separate description-page revision. These dates describe different things. Cached projected metadata can be up to five minutes old, and later edits can change source credits or files; recheck the selected source before publishing.

What a useful result looks like

A concise comparison table: File: title → source page → dimensions and format → recorded creator/credit → license and attribution terms or missing fields → file version → description-page revision → fetchedAt/cacheHit. Add the subject fit and unresolved source questions after human visual review.

Know the limits

Two read-only metadata tools, with no external provider key or paid provider. Hosted calls require a ToolCargo account and scoped key or OAuth and use the shared plan allowance, including cache hits. Search pages are explicit and bounded to ten candidates and offset 1000. No image preview, download, historical-file browsing, link following or legal clearance. Wikimedia Commons provides the source and the Wikimedia Foundation operates its public API; upstream availability is not guaranteed.