AI agent collects 471 product reviews in 29 min and clusters complaints into themes

eyishazyer · x · 2026-11-05

Developer eyishazyer shows how to turn a product's public reviews into one spreadsheet with an AI agent — using Meta Ray-Ban Display as the test case with the Oxylabs Web API MCP server: 471 reviews in 29 minutes.

Setup: install oxylabs/web-api-mcp via uv, add the web-api-skills, plug in an API key; ran Opus 5.5 in Claude Code (Cursor/Codex work too).
One prompt: find public review pages for Meta Ray-Ban Display, skip login walls, collect all reviews from the last 12 months, dedupe, save to reviews.csv.
Execution: 12 US-scoped searches, 57 scrape calls covering every Best Buy listing page plus Reddit; 25 calls pulled just the needed fields as JSON — 548 Best Buy reviews in, 462 kept, plus 9 Reddit posts.
Second prompt, no new web calls: group complaints into 5–8 themes with counts, percentages, and exact quotes with links; results in 4 min: limited apps 23%, weight and fit 15%, battery 12%.
Verification: every quote links back to its source review, and the agent itself flagged that all Best Buy reviews carried an "incentivized" badge.

Why Oxylabs: country-targeted search, markdown output agents can read directly, named-field JSON extraction, JavaScript rendering, and no charge for failed requests.

Original post →

AI agent collects 471 product reviews in 29 min and clusters complaints into themes

More from coding & agent