{
  "markdown": "# Offers MCP\n\n**Every offer. One call.**\n\nDeep purchase intelligence for AI agents, applications, and humans. Point\nit at a product (free text like `Sony WH-1000XM6 black`, or a GTIN/UPC/EAN\nbarcode), add an optional US zip code, and one call scans merchants in\nparallel to return live prices, true cost, delivery options, and trust\ninsights. Robust, private, efficient results for agents, applications, and humans to make\ninformed purchase decisions.\n\n**Robust. Built to be trusted:**\n\n- **Live prices with true landed cost** (price + shipping, the number that actually matters)\n- **Delivery windows for a given zip code**\n- **Merchant trust scores** (no pay-to-rank), with untrusted listings filtered\n- **Confidence scores** on the product match and on every offer, so your agent knows when to assert and when to hedge. When we're not sure, the response says so instead of guessing\n- A computed **best trusted offer** and **fastest trusted offer**, plus the full option list. The final call stays with your user\n- **Provenance on everything:** every offer names its source and the moment it was checked (`checked_at`, cache status)\n\n**Private. Nobody watches users shop:** no per-user tracking, no raw IPs\nstored; queries stay private: never sold, shared, or used for retargeting\n(details in [Privacy](#privacy)).\n\n**Efficient. Shop many stores in seconds:** one input fans out to\nmultiple sources at once and comes back normalized into one structured\ncontract you can build against. Repeat lookups return instantly. Use it for a\ncross-vendor price check from the store aisle or programmatic lookups inside\nyour own application. Same tool call, same answers.\n\nFree during v0. No signup. No API key. No commitment.\n\n## Quick start\n\nThe hosted server speaks streamable HTTP at:\n\n```\nhttps://offers-engine.dexter-works.com/mcp\n```\n\n### Claude Code\n\n```bash\nclaude mcp add --transport http offers https://offers-engine.dexter-works.com/mcp\n```\n\n### Claude Desktop / claude.ai (no terminal needed)\n\n1. Open **Settings** → **Connectors** (paid plans).\n2. Click **Add custom connector**.\n3. Name it `Offers`, paste the URL `https://offers-engine.dexter-works.com/mcp`, and save.\n4. In a new chat, make sure Offers is enabled in the tools/connectors menu, then just ask your shopping question.\n\n### ChatGPT\n\n1. **Settings** → **Apps & Connectors** → **Advanced settings** → enable **Developer mode** (Plus/Pro/Team plans).\n2. Back in **Connectors**, choose **Create** / add a custom connector.\n3. Name it `Offers`, MCP server URL `https://offers-engine.dexter-works.com/mcp`, authentication: **none**.\n4. In a chat, enable the Offers connector from the tools menu and ask away.\n\n### Cursor\n\nAdd to `.cursor/mcp.json` (project) or `~/.cursor/mcp.json` (global):\n\n```json\n{ \"mcpServers\": { \"offers\": { \"url\": \"https://offers-engine.dexter-works.com/mcp\" } } }\n```\n\n### VS Code (GitHub Copilot)\n\nAdd to `.vscode/mcp.json`:\n\n```json\n{ \"servers\": { \"offers\": { \"type\": \"http\", \"url\": \"https://offers-engine.dexter-works.com/mcp\" } } }\n```\n\n### Windsurf\n\nAdd to `~/.codeium/windsurf/mcp_config.json`:\n\n```json\n{ \"mcpServers\": { \"offers\": { \"serverUrl\": \"https://offers-engine.dexter-works.com/mcp\" } } }\n```\n\n### Any stdio-only MCP client\n\nUse [`mcp-remote`](https://www.npmjs.com/package/mcp-remote) as a bridge:\n\n```json\n{\n  \"mcpServers\": {\n    \"offers\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"mcp-remote\", \"https://offers-engine.dexter-works.com/mcp\"]\n    }\n  }\n}\n```\n\n> Client UIs move fast. If a menu path above is stale,\n> [open an issue](https://github.com/dexterworks/offers-mcp/issues) and we'll fix it.\n\nThen ask your assistant something like:\n\n> Use Offers MCP to check if I'm getting a good deal on the Sony WH-1000XM6 at $398. I'm in 83702.\n\nNaming the tool matters: assistants have their own web search and will often\nreach for it first on shopping questions. Saying \"use Offers MCP\" (or enabling\nonly this connector for the chat) makes sure your question hits live\nstructured offers instead of search snippets.\n\n## REST API\n\nThe same engine is available over plain HTTP:\n\n```bash\ncurl \"https://offers-engine.dexter-works.com/api/search_offers?q=Sony%20WH-1000XM6%20black&zip=83702\"\n```\n\n## Tools\n\n| Tool | Input | Output |\n|---|---|---|\n| `search_offers` | `query` (text or GTIN), `zip` (optional, 5-digit US) | Normalized offers + best/fastest trusted offer |\n\nMore tools (`identify_product` for GTIN/URL/image resolution and variant\nfamilies) are on the roadmap.\n\n## Response shape (abridged)\n\n```json\n{\n  \"product\": { \"name\": \"Sony WH-1000XM6…\", \"brand\": \"Sony\", \"resolution_confidence\": 0.9 },\n  \"offers\": [\n    {\n      \"merchant\": { \"name\": \"Best Buy\", \"trust\": { \"score\": 95 } },\n      \"price\": { \"amount\": 398.0, \"list_price\": 449.99, \"discount_pct\": 11.6 },\n      \"fulfillment\": {\n        \"landed_cost\": 398.0,\n        \"delivery\": { \"zip\": \"83702\", \"min_days\": 2, \"max_days\": 2, \"basis\": \"merchant_stated\" }\n      },\n      \"variant_match\": { \"confidence\": 0.98 },\n      \"provenance\": { \"checked_at\": \"2026-08-11T19:42:03Z\", \"cache\": \"miss\" }\n    }\n  ],\n  \"summary\": {\n    \"best_offer_id\": \"dw:off_…\",\n    \"recommendation\": \"Best Buy at $398.00 with free shipping is the best trusted deal.\"\n  }\n}\n```\n\n## Sources, limits, and expectations\n\n- **Sources (quality data, not screen-scraping):** we work through official\n  merchant and catalog partner programs and structured data sources so\n  results stay accurate rather than brittle. Today offers come from Google\n  Shopping, with retailer-direct integrations (Best Buy, Walmart) rolling\n  out; GTIN inputs are resolved via Icecat and UPCitemdb. Each response\n  lists exactly what was queried in `meta.sources_queried`, and where\n  sources disagree or fall short, the response says so.\n- **Latency:** a cold query takes roughly 5–15 seconds (live upstream\n  lookups); repeated queries hit a short-lived cache and return in\n  milliseconds. Configure generous tool timeouts.\n- **Rate limits / fair use:** free during v0. About **10 fresh lookups per\n  visitor per day** (cached queries are unlimited and don't count); over-cap\n  requests get an honest capacity message, and capacity resets daily (UTC).\n  Heavy or commercial use: contact us first. The endpoint may evolve;\n  breaking changes will be announced in this repo.\n- **Agent contract:** machine-readable guidance for agents (how to interpret\n  confidence, freshness, trust, and affiliate fields) lives at\n  [offers-engine.dexter-works.com/agents.md](https://offers-engine.dexter-works.com/agents.md).\n- **Feedback tool:** `report_feedback(query, verdict, note?)` files a result\n  report straight from your assistant; every report is read.\n- **REST and MCP return the same JSON shape** (the MCP tool wraps it as text\n  content).\n- **Note when curling `/mcp` directly:** responses are SSE-framed\n  (`event: message` / `data: {...}`) per MCP streamable HTTP; MCP clients\n  handle this for you.\n\n## Privacy\n\nWe built Offers for ourselves first, and we treat your searches the way we'd\nwant ours treated.\n\n- **What we see:** the product you searched, the zip code you optionally\n  provide, and coarse technical context (which kind of client connected,\n  country/region). We don't track individuals or build per-user shopping\n  profiles, and anonymous use requires no account at all.\n- **What we don't keep:** raw IP addresses are never stored; unique usage is\n  counted with a hash that changes daily and can't be reversed.\n- **What it's used for:** one thing: making results better (match accuracy,\n  merchant coverage, speed).\n- **What we will never do:** sell your data, share it with advertisers or\n  data brokers, or use queries for retargeting. Full stop.\n- **Merchants can't watch you compare.** All price comparison happens on our\n  servers; a merchant learns about a shopper only at click-through to their\n  site to buy. Some outbound links may earn us an affiliate commission at no\n  cost to you, and commissions never influence ranking (trust scores and landed\n  cost decide, and that logic treats every merchant the same).\n\nQuestions or want something deleted? [hello@dexter-works.com](mailto:hello@dexter-works.com).\n\n## Status & support\n\nEarly v0: US consumer goods, a small set of sources done honestly. If a\nfield can't be trusted, it's `null` rather than guessed.\n\nFound a bad result, wrong price, or a bug? **Open a\n[GitHub issue](https://github.com/dexterworks/offers-mcp/issues)**. Bad-match\nreports with the exact query are especially valuable right now.\n\nBuilt by [Dexter Works](https://dexter-works.com).\n",
  "bytes": 8547,
  "sha": "778271dd7b978c4a0aff499e8ce5e67fe0d8a12225d9db9e8069a2eb236428d2",
  "repo_slug": "dexterworks/offers-mcp",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_com_dexter_works_offers_2bdeb87d/readme"
}