SFor builders

The same evidence, for your agent.

Everything SafeShelf shows a shopper, your agent can ask for: over MCP, over REST, or agent to agent with A2A. Every tool is read-only. An agent can check a listing; it can never buy, report or act for a shopper.

Connect over MCP API docs

> is amazon.in/dp/B0CX65SFC safe to buy?

Ten read-only tools at /mcp.

Point any MCP client at the server and send your key in the x-api-key header. Answers carry evidence ids your agent can show the shopper.

check_listingREAD-ONLY

"Is it real?": copied photos, an impossible price, a brand-new website, conflicting specs, hidden AI instructions.

identify_productREAD-ONLY

Which product is this listing? Returns a SafeShelf product_id for the other tools. Give a url or a title.

product_safetyREAD-ONLY

Official recall status, hazard reports found in reviews and photos, and safety-mark checks.

find_copiesREAD-ONLY

The same product in other listings or brands: exact model, possible rebrand, variant or look-alike, with reasons.

safety_marksREAD-ONLY

Safety and energy marks the listing claims, checked against the official registries.

safer_alternativesREAD-ONLY

Similar products with no recall and no hazard reports, with the reason for each.

owner_guidanceREAD-ONLY

For someone who already owns it: stop or take care, the recall remedy, return, or report.

product_historyREAD-ONLY

Everything that ever changed about the product, and its recalls.

case_fileREAD-ONLY

Findings with evidence ids and quotes, and the code rule behind the verdict.

investigation_traceREAD-ONLY

Every step of an investigation: tool, model, time, tokens and estimated cost.

Or call the API directly.

The API runs on Cloud Run in Mumbai (asia-south1). The live investigation streams its steps as server-sent events, so your agent can show progress as it arrives.

EndpointsREST and A2A

Endpoint	What it does
POST/v1/agent/sessions/stream	Live investigation, streamed step by step
POST/v1/agent/sessions	Chat about a product, one answer
GET/.well-known/agent-card.json	A2A agent card
GET/health	Is the API up

Request fields and every response shape are in the API docs.

Check a listing

curl -N -X POST \
  https://safeshelf-api-368488022657.asia-south1.run.app/v1/agent/sessions/stream \
  -H 'content-type: application/json' \
  -H "authorization: Bearer $FIREBASE_ID_TOKEN" \
  -d '{"mode": "investigate", "product_id": "unmatched",
       "listing": {"title": "65W Super Fast Charger CX-65",
                   "url": "amazon.in/dp/B0CX65SFC",
                   "price": 299, "currency": "INR"}}'

MCP client config

{
  "mcpServers": {
    "safeshelf": {
      "url": "https://safeshelf-api-368488022657.asia-south1.run.app/mcp",
      "headers": { "x-api-key": "YOUR_KEY" }
    }
  }
}

Build on the evidence. Never on a guess.

Read-only tools, sourced answers, and a verdict set by code.

Read the API docs View the source