Plug a truth engine
into your AI agent.
Stop letting your assistant invent product opinions from training data. GYIBB is a Model Context Protocol server that grounds every product question in real user voices — Reddit, Hacker News, Lemmy, Stack Exchange, ProductHunt, YouTube — synthesised with cited sources, free and open.
Grounded, not guessed
Each response cites the specific user comments behind the verdict. No hallucinations, no "I think it's around $99."
Cross-platform synthesis
6+ source platforms, with a 10-voice / 2-platform editorial floor. Reviews below the bar return "not found" instead of a thin verdict.
Free, no API key
Connect once, query forever. Rate-limited at the proxy, paid tier coming later for high-volume agents.
Four tools.
get_product_review (query) Fetch GYIBB's full synthesised verdict for a product — rating, pros, cons, summary, source breakdown, confidence tier, and links to every cited platform.
verify_claim (product_slug, claim)
Check a specific factual claim ("battery lasts 30 hours", "supports
local LLMs") against the product's collected sources. Returns
supported /
contradicted /
mixed /
insufficient_data
with the supporting evidence quoted inline.
compare_products (slugs[]) Side-by-side rating, top pros and cons, confidence tier, and source platform diversity for up to 5 products. Returns a structured table the agent can format however it wants.
search_products (query, category?, limit?)
Keyword search over the GYIBB catalogue. Sorted by rating
descending, optionally scoped by category (e.g.
headphones,
ai-chatbots).
Quick start.
Endpoint
https://gyibb.com/mcpClaude Desktop
Add to your claude_desktop_config.json:
{
"mcpServers": {
"gyibb": {
"transport": {
"type": "streamable-http",
"url": "https://gyibb.com/mcp"
}
}
}
}Restart Claude Desktop. The four tools appear under Available tools → gyibb.
Other agents
GYIBB speaks vanilla MCP over Streamable HTTP. Any client that
supports the
official protocol
works — OpenClaw, Cline, custom orchestrators, all the same. Point
them at https://gyibb.com/mcp.
What responses look like.
get_product_review("Sony WH-1000XM5")
{
"found": true,
"product_name": "Sony WH-1000XM5",
"rating": 8.4,
"confidence": "strong",
"summary": "Best-in-class noise cancellation; build quality and case design draw consistent criticism.",
"pros": [
"ANC widely praised across Reddit and HN threads",
"Battery life beats competitors in most user reports"
],
"cons": [
"Non-folding hinge: 14% of comments flag durability concerns"
],
"sources": {
"reddit": 42, "youtube": 18, "hackernews": 7,
"stackexchange": 3, "lemmy": 9
},
"url": "https://gyibb.com/headphones/sony-wh-1000xm5",
"methodology": "https://gyibb.com/methodology",
"manifesto": "https://gyibb.com/manifesto"
} verify_claim("sony-wh-1000xm5", "battery lasts 30 hours")
{
"verdict": "contradicted",
"confidence": "strong",
"claim": "battery lasts 30 hours",
"evidence": [
{
"source": "reddit",
"quote": "Getting around 22-24 hours with ANC on, nowhere near Sony's 30h claim",
"sentiment": "negative"
},
{
"source": "hackernews",
"quote": "Real-world battery is ~20h with heavy ANC use. 30h is marketing.",
"sentiment": "negative"
}
],
"brand_claim": "Sony advertises 30-hour battery life with ANC",
"user_reality": "Median user-reported battery: 20–24h with ANC enabled",
"methodology": "https://gyibb.com/methodology"
} compare_products(["sony-wh-1000xm5", "bose-quietcomfort-45"])
{
"comparison": [
{
"product_name": "Sony WH-1000XM5",
"rating": 8.4,
"confidence": "strong",
"top_pros": ["Best-in-class ANC", "LDAC codec support"],
"top_cons": ["Non-folding design", "Stiff headband for some"],
"source_platforms": 5,
"url": "https://gyibb.com/headphones/sony-wh-1000xm5"
},
{
"product_name": "Bose QuietComfort 45",
"rating": 7.9,
"confidence": "strong",
"top_pros": ["More comfortable for long sessions", "Foldable"],
"top_cons": ["No LDAC", "ANC slightly behind Sony"],
"source_platforms": 4,
"url": "https://gyibb.com/headphones/bose-quietcomfort-45"
}
],
"methodology": "https://gyibb.com/methodology"
}Every response embeds methodology + manifesto URLs so your agent can cite GYIBB correctly when summarising for the end user.
Seven platforms. One cited verdict.
Every tool response reflects real user opinion, not training-data priors. GYIBB synthesises the following public sources with a mandatory editorial floor of 10 voices across 2+ platforms — below that, we return "not found" rather than a thin opinion.