GoBuy Product Trust MCP Server
Official# GoBuy Product Trust MCP Server
A public, read-only MCP server that gives AI agents GoBuy marketplace-evidence scores — and the reference evaluator for the **GPV-1 Pre-Purchase Verification Specification**.
Public endpoint: `https://mcp.gobuy.ai/mcp` — keyless, no authentication, read-only. Service statistics: `https://mcp.gobuy.ai/stats`.
## Tools (5)
- **check_product_trust** `{retailer, product_id}` — inspect one Amazon, Walmart, Target, or Best Buy listing: Evidence Score (0-100), verdict, observed/missing signals, freshness. This is the GPV-1 gate evaluator (score >= 55 PASS - 35-54 FLAG - < 35 or not-indexed BLOCK).
- **compare_products** `{products: [...]}` — inspect and rank 2-20 listings by Evidence Score.
- **analyze_reviews** — per-review forensics: review integrity score, adjusted rating, suspicious-review flags, Verified Purchase checks.
- **check_store_score** `{domain}` — GoBuy Store Score for any storefront (Shopify and others): agent-readiness plus trust signals, tiers A-D.
- **scan_store** `{domain}` — on-demand deep storefront scan.
Example:
```json
{ "retailer": "amazon", "product_id": "B0BXYCS74H" }
```
## GPV-1 — Pre-Purchase Verification
This server's `check_product_trust` is the reference evaluator for [GPV-1](https://docs.gobuy.ai/standard), the open pre-purchase verification specification: eight rules (R1-R8) that must all be true before an agent proceeds to checkout. Machine-readable ruleset: [docs.gobuy.ai/standard/v1.json](https://docs.gobuy.ai/standard/v1.json). Reference integration: [verify-then-checkout](https://github.com/digitalmentes7-maker/verify-then-checkout) (Next.js boilerplate - a secured shopping agent in 5 minutes).
## Evidence disclaimer
Evidence Scores describe available marketplace records, not product quality, safety, value, suitability, seller legitimacy, or a purchase recommendation. Retailers do not endorse GoBuy.
## Client configuration
### Any MCP Client
```json
{ "mcpServers": { "gobuy-product-trust": { "url": "https://mcp.gobuy.ai/mcp" } } }
```
### Claude Desktop
```json
{ "mcpServers": { "gobuy-product-trust": { "command": "npx", "args": ["-y", "mcp-remote", "https://mcp.gobuy.ai/mcp"] } } }
```
### Inspector
`npx @modelcontextprotocol/inspector` -> Streamable HTTP -> `https://mcp.gobuy.ai/mcp` -> list tools -> call `check_product_trust`.
## Transports
- **stdio** (for `mcp-proxy` bridges, e.g. Glama deploys): set `MCP_TRANSPORT=stdio`. The server speaks MCP over stdin/stdout; stdout carries only JSON-RPC, all logs go to stderr.
- **HTTP** (default): listens on `PORT` (default 3000), endpoint `/mcp`, health at `GET /mcp`, stats at `GET /stats`.
## Limits
No client authentication required. Rate limited per IP: 60/minute, 1,000/day (429 + Retry-After). Everything is free within rate limits; an optional free API key raises them.
## License
MIT
## Changelog
- 2026-10-14: stdio transport (MCP_TRANSPORT=stdio) for mcp-proxy deployments; stdout reserved for JSON-RPC, logs to stderr.
TDQS
Scored across 2 tools
The two tools have clearly distinct purposes: check_product_trust scores a single retailer listing, while compare_products ranks 2-20 listings against each other. The descriptions make the singular-vs-multi scope explicit, so an agent cannot reasonably confuse them.
Both names follow the same predictable verb_noun snake_case pattern (check_product_trust, compare_products). The leading verbs (check, compare) also accurately signal the differing action.
Two tools is on the thin side, but the domain (scoring and comparing retailer listings) is narrow enough that each tool earns its place. A slightly broader surface (e.g. batch or signal-discovery) would round it out.
The core workflow of inspecting one listing and comparing several is covered, which matches the stated purpose. Minor gaps exist: no batch check for a list of single products, no operation to discover or list available signals/retailers independently.