GYIBB Truth Engine — MCP Server

Cited product-review verdicts synthesised from real user voices. Free, remote, no API key.

Endpoint: https://gyibb.com/mcp
Transport: streamable-http
Auth: none
Docs for agents: https://gyibb.com/for-agents

Most product "reviews" on the web are marketing copy, affiliate filler, or incentivised ratings. GYIBB reads what actual users said on Reddit, Hacker News, Lemmy, Stack Exchange, ProductHunt, YouTube and Trustpilot, and synthesises it into a structured verdict an agent can act on — with the raw quotes still attached.

Every claim traces back to a source. A product without 10+ user voices across 2+ platforms is refused rather than written up from thin air: an empty answer beats a confident guess.

Connect

Claude Desktop

Add to claude_desktop_config.json:

config.json

{
  "mcpServers": {
    "gyibb": {
      "type": "streamable-http",
      "url": "https://gyibb.com/mcp"
    }
  }
}

Any MCP client

Point it at https://gyibb.com/mcp over streamable-http. No signup, no key, no OAuth. If your client only speaks stdio, bridge it with mcp-remote:

Terminal

npx -y mcp-remote https://gyibb.com/mcp

Verify it's up

Terminal

curl -N -X POST https://gyibb.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"curl","version":"1"}}}'

A healthy server answers 200 with content-type: text/event-stream and a serverInfo.name of gyibb-truth-engine. The response is an SSE stream, so use -N — without it curl waits for a close that never comes and looks like a timeout.

Tools

Tool	Arguments	Returns
`get_product_review`	`query` — product name, slug, or `category/slug`	Full verdict: rating, confidence tier, pros, cons, summary, source breakdown, citation links
`verify_claim`	`product_slug`, `claim`	`supported` / `contradicted` / `mixed` / `insufficient_data`, with the quotes behind the call
`compare_products`	`slugs` — up to 5	Side-by-side ratings, top pros/cons, confidence, source diversity
`search_products`	`query`, `category` (optional), `limit` (optional)	Rating-sorted matches from the catalogue

verify_claim is the one that has no obvious equivalent elsewhere: it answers "is this marketing claim true according to people who bought it", and returns insufficient_data instead of inventing a verdict when the evidence isn't there.

Rate limits

	`tools/call` per IP per UTC day
Anonymous	25
With public key	100

initialize, tools/list and notifications are free.

To raise the quota, send the public integration key — it is not a secret, it exists to separate real integrations from drive-by traffic:

Code

X-API-Key: we-read-it-so-you-dont

What the verdicts are built from

Reddit · Hacker News · Lemmy · Stack Exchange · ProductHunt · YouTube · Trustpilot · editorial aggregate ratings.

Each review is a synthesis across four layers — what users report, what video reviewers show, what aggregate ratings say, and what the brand claims — and the most useful field is usually where those layers disagree. Reviews are written by AI agents against a published methodology, then cross-checked by an independent fact-checking agent before release.

Methodology: https://gyibb.com/methodology
Editorial stance: https://gyibb.com/manifesto

Files here

File	Purpose
`server.json`	Official MCP Registry manifest (`io.github.Patheras/gyibb-mcp`)
`.well-known/mcp.json`	Server manifest served for discovery — tools, limits, sources

The server itself is not in this repository; it is part of the GYIBB platform and served at https://gyibb.com/mcp. This repo is the public front door: manifest, registry entry, and these docs.

Licence

Free for read-only use. See https://gyibb.com/manifesto.

Contact: founder@gyibb.com