Swervie.
Sign in

MCP server for shopping agents

The whole market's product data,
at 1% of the tokens.

Swervie continuously extracts specs from retailers across the market at 95% measured accuracy, deduplicates identical products across sellers, and serves one ranked answer over MCP. A full shopping task costs your agent ~4,000 tokens and finishes in under a second - not 400,000 tokens of bot-walled HTML per retailer page.

Add to any MCP clientfree · no key
{
  "mcpServers": {
    "swervie": {
      "url": "https://us-central1-swervie-prod.cloudfunctions.net/mcp"
    }
  }
}
agent session · swervie mcp
› find_category({"query": "65 inch OLED TV for bright rooms"})
→ {"category": "Televisions", "products": 341, "dimensions": 42}
› browse_category({"category": "Televisions", "filters": [{"dimension": "Display Technology", "eq": "OLED"}]})
→ 7 consolidated groups, value-ranked. Best: {"title": "LG C3 65\" OLED", "sellers": 4, "best_price": "$1,299", "swervie_score": 91}
› get_products({"ids": ["swervie:group:lg-c3-65"]})
→ 42 verified specs, score receipts, 4 buy links. One answer, not ten tabs.

3 calls · 14.1 KB · 0.78s warm. The same task over raw retailer pages: ~400k tokens each, when the bot wall lets you in.

1,353

products tracked across 40+ retailers

86

cross-seller product groups, deduplicated by GTIN + model

95%

measured extraction accuracy, per-attribute provenance

42

typed spec dimensions on the deepest category schemas

What a shopping task costs your agent

Same live task on each channel: find the best 65" OLED with verifiable specs and a current price.

Swervie MCP

~4k tokens · 0.78s warm

3 calls. Deduplicated, spec-filtered, score receipts, per-seller buy links. Free, no key.

Agent + web search

1 search + 5–10 page fetches

Tens of seconds, unverifiable specs, duplicate listings - and we measured a years-expired "deal" presented as current.

Raw retailer pages

~400k tokens per page

100x the context per retailer, parsed by you - when the bot wall lets you in at all.

Four tools, one flow

find_category

Match a shopping intent to a category and its full spec schema - up to 42 typed dimensions.

browse_category

Deduplicated product groups, value-ranked, spec filters honored, current prices attached.

get_products

Complete spec sheets with per-attribute provenance, score receipts, per-seller buy links.

record_selection

Report what your human chose - ranking learns from it.

Also speaks UCP: catalog search + lookup at /api/ucp/mcp · agent card at .well-known/ucp-agent.json

Shopping as a human?

The full catalog - consolidated listings, spec comparisons, seller offers - is free with a Google sign-in.

Sign in to browse