An AI investigator for the honest web
Name a product, place, film, or company. The agent digs past page-one marketing to the candid, buried takes real people leave — Reddit, Trustpilot, forums — and files an honest verdict. With sources you can open.
Searched in real time on the live web via Nimble
The Archive
Real investigations the agent has already run. Hover to peek — open a file for the full dossier.
“It has a light that makes all the dirt show, but isn't powerful enough to pick it all up. I was really surprised at how weak the suction is.”
— reddit.com
“A failure as both spectacle and storytelling — pacing far too fast, writers jumbled Dune and Messiah into almost one.”
— reddit.com
“Quality control is terrible, even at their flagship — it's good at best, mid on average, terrible at worst.”
— reddit.com
“It sucks. It's using subpar models and unable to fulfill tasks such as identifying stuff inside specific databases when asked to.”
— reddit.com
“Sloppy content, outdated technology, high costs, and forced subscription issues noted after 5 years of ownership”
— reddit.com
The method
Not one API call — a research agent that plans, searches, critiques its own coverage, and digs again before it commits to a verdict.
Claude reads the subject and tailors search angles to it — Amazon reviews and Reddit for a product, Letterboxd and IMDb for a film, TripAdvisor and local subreddits for a restaurant, Glassdoor and G2 for a company.
Nimble runs every angle in parallel across the candid sources page one buries — and deep-reads the long-form angle, extracting full page text for the quotes that matter.
The agent reads what came back, finds the biggest blind spot, and writes a sharper follow-up query. This loop is what makes it an agent, not a search box.
It searches again to fill the gap — chasing the durability complaint, the repair thread, the thing no one advertises.
Claude synthesises an honest verdict and a trust score, citing sources by index so every quote links to a real page — never invented.
Ask ChatGPT about a product and it parrots the glossy surface — LLMs are blind to buried reviews. hidden.reviews is a hosted MCP server, so Claude, Cursor, or ChatGPT can pull the sourced, buried truth live — right inside your workflow.
https://hidden-reviews.vercel.app/api/mcpOr drop the snippet into your mcp.json
{
"mcpServers": {
"hidden-reviews": {
"url": "https://hidden-reviews.vercel.app/api/mcp"
}
}
}