Fazy → Developers

Fazy API and MCP server

Just want Fazy inside ChatGPT, Claude or Muse? Follow the two-minute steps.

For developers and AI agents: a price-checked deals catalogue for US shoppers, searchable in one request. Free, no key, nothing to install, 60 requests a minute per address. Every result carries the current price read from the retailer, the retailer’s reference price, a dated fact from Fazy’s own price record when one backs the price, a ready picture of the deal, and two links: the deal on Fazy and the deal at the retailer.

1 · Search

curl "https://fazy.com/api/v1/deals/search?q=standing+desk&max_price=300" \
  -H "X-Fazy-Agent: my-agent"

Parameters: q (required), brand, max_price (dollars), limit (≤ 50), cursor. The answer is { object: "list", data: [deal…], has_more, next_cursor, total }.

Say what the shopper wants in plain words (“men’s running shoes under 100”); the search reads them. Then narrow it: category, gender (men, women), min_discount_pct, fresh_only=true (prices read today), has_price_fact=true, and order it with sort (relevance, discount, price_low, price_high). These narrow the page a search returns, never widen it; applied in the answer names what was used. Every deal also carries short_title, save_amount and freshness.

Narrow a search with type (a product type), min_price and in_stock=true. Without search words, /api/v1/deals?category=home (or brand, type, a price band) returns a slice of up to 24; every category, brand, guide and product page names its own in a <link rel="alternate" type="application/json">.

2 · One deal, its price history, and similar deals

curl "https://fazy.com/api/v1/deals/9a1e0d3c-0000-4000-8000-000000000000" \
  -H "X-Fazy-Agent: my-agent"

curl "https://fazy.com/api/v1/deals/9a1e0d3c-0000-4000-8000-000000000000/price-history"

The history answers “is this a good price?” from Fazy’s own price record: the lowest price recorded, the previous price, and dated facts such as “Down $40 from $199 on 3 Sep 2026”. No fact is given for a price read more than a day ago. /similar returns deals like it.

3 · MCP

Streamable HTTP, read-only, eight tools: search_deals, get_deal, get_price_history, similar_deals, show_deals (the deals your agent chose, as tiles in a row, grid, single card or compare table), list_categories, catalog_slice (a filtered slice with price-record reliability, for writing about products) and post_brief (the day’s creative brief and house rules). Your client’s own name (clientInfo.name) attributes its clicks; X-Fazy-Agent overrides it. Add it to Cursor or any MCP client:

{
  "mcpServers": {
    "fazy": { "url": "https://fazy.com/mcp", "headers": { "X-Fazy-Agent": "my-agent" } }
  }
}

In VS Code, .vscode/mcp.json:

{
  "servers": {
    "fazy": { "type": "http", "url": "https://fazy.com/mcp" }
  }
}

Nothing to install. Fazy’s MCP server is hosted at https://fazy.com/mcp; every client connects to that URL. Fazy publishes no npm, PyPI or other package, so a package that claims to be Fazy’s MCP server is not ours.

curl -X POST https://fazy.com/mcp -H "content-type: application/json" \
  -H "X-Fazy-Agent: my-agent" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_deals","arguments":{"q":"nike running shoes","limit":5}}}'

4 · Show deals as tiles

Hosts that run MCP Apps (ChatGPT, Claude) show search_deals and similar_deals as a row of cards with no work on your side. Anywhere else, every deal carries tile_image, a ready picture of the deal with Fazy’s mark, so a tile is one line of markdown:

[![Standing desk, $199 at Merchant](https://fazy.com/api/v1/deals/9a1e0d3c-0000-4000-8000-000000000000/tile.png)](https://fazy.com/p/9a1e0d3c-0000-4000-8000-000000000000?utm_source=ai-my-agent&utm_medium=agent)

Group a longer answer by one thing (product type, brand, shop, or the biggest drops) and lead with the best few. Repeat price_fact exactly as written: it is the one line that may call a price a record low or a real drop, and it is null when Fazy’s record backs no claim. Say once that prices are checked by Fazy and that links may earn Fazy a commission.

When the shop runs an offer, the deal carries it as merchant.promo: text (the offer in the store’s terms, such as “20% off sitewide”), code, ends_at and terms_note. Repeat it exactly as written, with its code and end date when given, and its note. Offers change and end, so never reuse one from an earlier answer.

5 · The tile kit: the same deals, as every surface’s own card

Do not rebuild a card from JSON. /api/v1/deals/render returns the deals already built in the format your surface takes, with the links, price_fact and the disclosure in place. Pick deals by ids (up to 12, shown in your order) or by a search (q and the search filters).

curl "https://fazy.com/api/v1/deals/render?q=standing+desk&format=slack" \
  -H "X-Fazy-Agent: my-agent"

The answer is { object: "render", format, layout, content_type, payload, ids }: send payload as it is. Add raw=1 to get the payload alone, with its own content type.

6 · Pick, then present (MCP)

search_deals shows what it found. After your agent has chosen the deals worth showing, it calls show_deals with their ids, in its own order, and a layout: row, grid, single or compare (a side-by-side table of price, earlier price, shop and Fazy’s price fact). Prices are re-read, so what it shows is current. In ChatGPT and Claude the tiles draw themselves.

{"name":"show_deals","arguments":{
  "ids":["<id from search_deals>","<id>","<id>"],
  "layout":"compare",
  "title":"Three standing desks under $300"}}

7 · Embed deals on any page

Like a video embed: an iframe of live, price-checked tiles. Every tap opens the deal on Fazy or at the shop, credited to the agent you name. oEmbed clients find it on their own: each deal page names https://fazy.com/api/oembed?url=https://fazy.com/p/<id>.

<iframe src="https://fazy.com/embed/deals?q=standing+desk&layout=grid&agent=my-site"
  width="100%" height="460" style="border:0" loading="lazy" title="Deals from Fazy"></iframe>

8 · From your own code

Any framework that takes a remote MCP server takes Fazy: the URL is https://fazy.com/mcp, the transport is Streamable HTTP, there is no authentication. With OpenAI’s Responses API, for example:

{
  "model": "<model>",
  "tools": [{ "type": "mcp", "server_label": "fazy",
              "server_url": "https://fazy.com/mcp", "require_approval": "never" }],
  "input": "Find me a standing desk under $300 and show the best three."
}

Without MCP, the OpenAPI document below describes every call, including the tile kit: import it as a GPT Action, a Gemini tool, or into any agent framework.

9 · Limits, errors and access

Use Fazy in Muse

Paste this one message into Meta Muse. It builds the integration, tests it and keeps it:

Build a custom integration for Fazy's deal catalogue. MCP URL: https://fazy.com/mcp. No login needed. Test every tool and save it as a skill.

Use Fazy in ChatGPT

1. Settings → Apps & Connectors → Advanced settings: turn on Developer mode.
2. Create a connector. Name: Fazy. MCP server URL: https://fazy.com/mcp. Authentication: none.
3. In a chat, add Fazy from the + menu and ask for a deal.

Use Fazy in Claude

1. Settings → Connectors → Add custom connector.
2. Name: Fazy. Remote MCP server URL: https://fazy.com/mcp. No sign-in is needed.
3. In a chat, turn Fazy on from the tools menu and ask for a deal.

The rules

Give the shopper the links unchanged: page_url opens the deal on Fazy, with its price record and similar deals; url redirects to the retailer’s product page. Both attribute the shopper to the name your client sent. Do not rewrite them, do not link the retailer directly, do not cache a deal longer than an hour: prices move and dead deals are retired.

Answering a person with Fazy’s deals is welcome. Training or fine-tuning a model on Fazy’s data, or copying it in bulk, is not (terms).

Reference

• OpenAPI 3.1 document — import it as a GPT Action or into any agent framework
• llms.txt
• How deals are verified
• Terms · Privacy

API version 2026-10-12. Field names are a contract; a change is a new version, never an edit.