# Findcheap MCP server

> The MCP server reference, as Markdown for agents. The same page for people: https://findcheap.ai/developers/mcp. Index: https://findcheap.ai/llms.txt. Every answer's fields: https://findcheap.ai/developers/docs.md.

## Overview

The Findcheap MCP server lets an AI assistant find where a product is sold for less. It offers the capabilities of the Findcheap API as four tools that any MCP client (Claude, Claude Code, Cursor and others) can call in a conversation:

1. **find_cheaper** - takes a product page URL as input and searches the internet to find where the product is available for a lower price.
2. **read_product_page** - takes a product page URL as input and returns structured product data from the page.
3. **resolve_listing** - a utility tool that turns a result's `resolve_id` into the merchant's link and name.
4. **find_cheaper_feedback** - records how a search went. The first report on a search gives 1 credit back.

Server URL (Streamable HTTP): `https://mcp.findcheap.ai/mcp`

Each tool answers exactly what the corresponding API endpoint answers. This page covers how to connect and what each tool takes; every field of the answers is documented in the API reference. The same reference says how the API and these tools are versioned and lists every change, under Versioning and changes.

The examples on this page use our sample product page, https://findcheap.ai/sample-product: a copy of a real product page that does not change. Any product page's address works the same way.

## Connect

Two ways in, the same account behind both: sign in with your Findcheap account from ChatGPT, Claude.ai, Cursor or any other app that can sign in, or give a developer tool your API key.

### ChatGPT: sign in

Findcheap is a plugin in ChatGPT. Open its page there and add it:

[Open in ChatGPT](https://chatgpt.com/plugins/plugin_asdk_app_6afa139e1fec81918f9f5d115c4e28aa)

Then sign in: ChatGPT sends you to findcheap.ai, you sign in (or create an account, free) and allow ChatGPT to use it. From then on ChatGPT searches on your plan's credits.

### Claude.ai: sign in

In Claude (web, desktop or mobile) open Customize, then Connectors, and add a custom connector: name it Findcheap, enter the server URL below and continue. Under OAuth client choose "Use your own OAuth client" and paste this client ID (there is no secret):

- Server URL (Streamable HTTP): `https://mcp.findcheap.ai/mcp`
- OAuth client ID for Claude.ai: `d125ba2b-831c-403b-a456-a1141ef32231`

Add it, then sign in: Claude sends you to findcheap.ai, you sign in (or create an account, free) and allow Claude to use it. From then on Claude searches on your plan's credits.

### Cursor: sign in

Press the button and Cursor asks to add the server:

[Add to Cursor](cursor://anysphere.cursor-deeplink/mcp/install?name=findcheap&config=eyJ1cmwiOiJodHRwczovL21jcC5maW5kY2hlYXAuYWkvbWNwIiwiYXV0aCI6eyJDTElFTlRfSUQiOiIzZTEwZDA2Mi1jMzY2LTRmOGMtYTg1ZC1iODczOTUxNzg3OTMifX0=)

Or add it by hand:

```json
// .cursor/mcp.json (or the global ~/.cursor/mcp.json)
{
  "mcpServers": {
    "findcheap": {
      "url": "https://mcp.findcheap.ai/mcp",
      "auth": { "CLIENT_ID": "3e10d062-c366-4f8c-a85d-b87395178793" }
    }
  }
}
```

Then sign in: Cursor sends you to findcheap.ai, you sign in (or create an account, free) and allow Cursor to use it. From then on Cursor searches on your plan's credits.

### Any other app: sign in

Any app that can sign in to an MCP server connects with the server URL alone, with no API key. Add the URL in the app:

Claude Code:

```bash
claude mcp add --transport http findcheap https://mcp.findcheap.ai/mcp
claude mcp login findcheap
```

Other apps:

```json
// the app's MCP settings file
{
  "mcpServers": {
    "findcheap": { "url": "https://mcp.findcheap.ai/mcp" }
  }
}
```

Then sign in: the app sends you to findcheap.ai, you sign in (or create an account, free) and allow the app to use it. From then on the app searches on your plan's credits.

### Developer tools: the API key

Claude Code, Cursor, the Claude API and your own code take the key the API uses (create one in your dashboard, https://findcheap.ai/signin?next=/dashboard/keys, free). It travels in the `Authorization` header of every request.

Claude Code:

```bash
claude mcp add --transport http findcheap https://mcp.findcheap.ai/mcp \
  --header "Authorization: Bearer fc_live_..."
```

Cursor:

```json
// .cursor/mcp.json (or the global ~/.cursor/mcp.json)
{
  "mcpServers": {
    "findcheap": {
      "url": "https://mcp.findcheap.ai/mcp",
      "headers": { "Authorization": "Bearer fc_live_..." }
    }
  }
}
```

Claude API:

```python
import anthropic

# The Messages API connects to the server on Anthropic's side:
# Claude calls the tools itself, in the middle of the request.
client = anthropic.Anthropic()
response = client.beta.messages.create(
    model="claude-opus-5-5",
    max_tokens=4096,
    betas=["mcp-client-2025-11-20"],
    mcp_servers=[{
        "type": "url",
        "url": "https://mcp.findcheap.ai/mcp",
        "name": "findcheap",
        "authorization_token": "fc_live_...",
    }],
    tools=[{"type": "mcp_toolset", "mcp_server_name": "findcheap"}],
    messages=[{
        "role": "user",
        "content": "Find these cheaper: https://findcheap.ai/sample-product",
    }],
)
print(response.content[-1].text)
```

Python (the official MCP client):

```python
# pip install mcp
import asyncio
import httpx2
from mcp import Client
from mcp.client.streamable_http import streamable_http_client

async def main():
    headers = {"Authorization": "Bearer fc_live_..."}
    async with httpx2.AsyncClient(headers=headers, timeout=120) as http:
        async with Client(streamable_http_client("https://mcp.findcheap.ai/mcp", http_client=http)) as client:
            result = await client.call_tool("find_cheaper", {
                "url": "https://findcheap.ai/sample-product",
            })
            for card in result.structured_content["same_product"]:
                print(card["retailer"], card["price"], card["currency"], card["url"])

asyncio.run(main())
```

Any other MCP client connects the same way: the server URL above, Streamable HTTP, and the `Authorization` header.

## Authentication

Every request carries either your API key or the sign-in token an app holds for you:

```
Authorization: Bearer fc_live_...
```

A request without a valid key or token is answered `401` before anything else happens, which a client reports as a failed connection; the answer also points a client at the server's OAuth metadata, so a client that can sign in knows where to. A key is valid for the API and the MCP server alike, a sign-in for the MCP server, and all of them draw on the same credits and the same rate limits: a `find_cheaper` call costs 10 credits, a `read_product_page` call 1, a `resolve_listing` call 1 and a `find_cheaper_feedback` call nothing (the first report on a search gives 1 credit back); a call that fails costs nothing, except a `find_cheaper` call that stops because the page is not a product page or the used search is refused, which is billed as a page read, 1 credit.

The sign-in is standard OAuth 2.1: the server publishes its metadata at `/.well-known/oauth-protected-resource/mcp`, the authorization server is Findcheap's own, an app registers itself (OAuth dynamic client registration, at the `registration_endpoint` in the authorization server's metadata), and PKCE is required. An app's return address is `https`, or `http` on the person's own computer (`localhost`, on any port).

## find_cheaper

Takes a product page URL as input and searches the internet for where the product is available for a lower price. It is the Search endpoint as a tool: the agent reads the product page, then searches the marketplaces of the shopper's market. Results are sorted into two arrays, each containing up to 10 results:

- `same_product` - listings that match the original product. Sorted by cheapest first.
- `similar_products` - acceptable equivalents of the original product. Sorted by cheapest first.

It has a maximum budget of 60 seconds, but usually completes much sooner. While it runs, the server sends progress notifications (reading the page, searching, comparing) to clients that display them.

Two ways it gets used:

- **User-initiated.** A shopper pastes a product link into their assistant and asks for it cheaper; the assistant calls `find_cheaper` and answers with the retailers, prices and links.
- **Agent-initiated.** An agent working on a task the user gave it, "buy me a soccer ball", finds a candidate ball on Amazon and calls `find_cheaper` on that page to see whether the same ball, or an equivalent, is sold for less elsewhere before recommending or buying it.

In both, there is nothing to build: the assistant decides when to call the tool from its description.

### Parameters

- `url` (string, required) - The product page's address (https), as the shopper sees it.
- `market` (string) - The shopper's market, the country they are buying in (see Markets in the API reference). If omitted, the market is inferred from the URL; otherwise, it defaults to US.
- `marketplaces` (string[]) - Which marketplaces to search. Only marketplaces available in the shopper's market count. If omitted, the agent chooses for the product. Accepted values: `"amazon"`, `"walmart"`, `"ebay"`, `"aliexpress"`, `"shein"`, `"flipkart"`, `"other_sites"`.
- `condition` (`"new"` | `"used"`) - Defaults to `"new"`. If set to `"used"`, the agent searches for where the product can be bought secondhand. This search mode is not available in India, and is refused for grocery and consumable products.
- `max_results` (integer) - How many listings at most in each list, `1` to `10`; 10 when omitted. Ask for fewer when only the cheapest few matter. The search is the same, and so are the credits.

### Result

The Search endpoint's answer, as JSON, in this order: `search` (what was searched: the address, the market, the condition, the cap), `product`, `product_name`, `about_lists` (what the two lists hold for this search), `same_product`, `similar_products`, `page_is_cheapest` and `next_steps` (what can be done next, each with the call that does it). The endpoint's `request_id` and `credits_used` are not in the answer; they are in the result's `_meta` (see The call's id and cost, below). Every field, and the structure of a card, is documented under Search in the API reference (https://findcheap.ai/developers/docs.md). A card whose `url` is `null` carries a `resolve_id` for `resolve_listing`. Each tool also publishes its answer's schema in the tool listing (`outputSchema`: every field, its type and whether it can be null), so a client can check an answer before it shows it.

### Example

The call the assistant makes, and the answer trimmed to one card per list:

```
find_cheaper({ "url": "https://findcheap.ai/sample-product" })
```

```json
{
  "search": {
    "url": "https://findcheap.ai/sample-product",
    "market": "US",
    "condition": "new",
    "max_results": 10
  },
  "product": {
    "title": "Nike Dunk High By You Custom Men's Shoes",
    "price": 160.0,
    "currency": "USD",
    "brand": "Nike",
    "url": "https://findcheap.ai/sample-product",
    "market": "US",
    "image": "https://findcheap.ai/demo-photos/nike-dunk-high-by-you.png",
    "in_stock": true,
    "retailer": null,
    "coupon": null,
    "promo_code": null,
    "shipping_cost": null,
    "returns_accepted": null,
    "return_days": null,
    "return_shipping_paid_by": null,
    "pack_count": null
  },
  "product_name": "Nike shoes",
  "about_lists": "same_product holds listings from the original product's brand that look the same or similar. similar_products holds look-alikes from any brand.",
  "same_product": [
    {
      "title": "Nike Dunk Hi Retro Blue/White Leather Men’s US 11 Sneaker DD1399-102",
      "price": 69.0,
      "currency": "USD",
      "retailer": "eBay",
      "url": "https://www.ebay.com/itm/178217128667",
      "resolve_id": null,
      "image": "https://i.ebayimg.com/images/g/VGwAAeSwwNtqLLg0/s-l500.jpg",
      "over_page_price": false,
      "rating": 100.0,
      "coupon": null,
      "promo_code": null,
      "price_before_discount": null,
      "discount_type": null,
      "discount_value": null,
      "shipping_cost": null
    }
  ],
  "similar_products": [
    {
      "title": "Nike By You Dunk High Sail Black FV5511 900 Men's Size 10.5",
      "price": 59.99,
      "currency": "USD",
      "retailer": "eBay",
      "url": "https://www.ebay.com/itm/354741245174",
      "resolve_id": null,
      "image": "https://i.ebayimg.com/images/g/BIAAAOSwl2RkRuKP/s-l500.jpg",
      "rating": 100.0,
      "coupon": null,
      "promo_code": null,
      "price_before_discount": null,
      "discount_type": null,
      "discount_value": null,
      "shipping_cost": null,
      "over_page_price": false
    }
  ],
  "page_is_cheapest": false,
  "next_steps": [
    {
      "label": "Find pre-owned",
      "tool": "find_cheaper",
      "arguments": { "url": "https://findcheap.ai/sample-product", "market": "US", "condition": "used" },
      "choose": null
    },
    {
      "label": "Filter stores",
      "tool": "find_cheaper",
      "arguments": { "url": "https://findcheap.ai/sample-product", "market": "US", "condition": "new" },
      "choose": { "marketplaces": ["amazon", "walmart", "ebay", "aliexpress", "shein", "other_sites"] }
    }
  ]
}
```

## read_product_page

Takes a product page URL as input and returns structured product data from the page, without searching. It is the Extract endpoint as a tool. It answers within 30 seconds, and has a median response time of 4 seconds.

Two ways it gets used:

- **User-initiated.** A shopper pastes a product link into their assistant and asks what it sells, what it costs or whether it is in stock; the assistant calls `read_product_page` and answers from the page's data.
- **Agent-initiated.** An agent working on a task the user gave it, "compare these three monitors", calls `read_product_page` on each page to get the title, price, brand and stock it needs before comparing them.

In both, it answers what the page says and nothing more; `find_cheaper` is the tool that searches.

### Parameters

- `url` (string, required) - The product page's address (https).
- `market` (string) - The market to read the page from (see Markets in the API reference). If omitted, the market is inferred from the URL; otherwise, it defaults to US.

### Result

The Extract endpoint's answer: `is_product_page`, `title`, `image`, `price`, `currency`, `brand`, `in_stock`, `retailer`, `coupon`, `promo_code`, `shipping_cost`, `returns_accepted`, `return_days`, `return_shipping_paid_by`, `url`, `market`, `reason` and `next_steps` (what can be done next for this product: the search for it, with the call that does it). Every field is documented under Extract in the API reference.

### Example

```
read_product_page({ "url": "https://findcheap.ai/sample-product" })
```

```json
{
  "is_product_page": true,
  "url": "https://findcheap.ai/sample-product",
  "market": "US",
  "title": "Nike Dunk High By You Custom Men's Shoes",
  "image": "https://findcheap.ai/demo-photos/nike-dunk-high-by-you.png",
  "price": 160.0,
  "currency": "USD",
  "brand": "Nike",
  "in_stock": true,
  "retailer": null,
  "coupon": null,
  "promo_code": null,
  "shipping_cost": null,
  "returns_accepted": null,
  "return_days": null,
  "return_shipping_paid_by": null,
  "reason": null,
  "next_steps": [
    {
      "label": "Find this cheaper", "tool": "find_cheaper",
      "arguments": { "url": "https://findcheap.ai/sample-product", "market": "US" },
      "choose": null
    }
  ]
}
```

## resolve_listing

For some results, `find_cheaper` only returns top-level data, to maximize search efficiency. These results carry a `resolve_id` instead of a `url`. Calling `resolve_listing` with a `resolve_id` returns a direct link to the corresponding product on the merchant's website and the merchant's name. It is the Resolve endpoint as a tool.

An id is valid for up to two hours after the search that produced it. In rare cases, it may expire sooner, so it is advised to resolve any desired ids shortly after the search has completed.

### Parameters

- `resolve_id` (string, required) - The card's `resolve_id`.

### Result

| Field | Type | Meaning |
| --- | --- | --- |
| `resolved` | boolean | true when url holds the retailer's link. |
| `url` | string \| null | The product's link on the merchant's website, tracking parameters stripped. null when not resolved. On the Free plan it may be an affiliate link; see Links in the API reference (https://findcheap.ai/developers/docs.md). |
| `retailer` | string \| null | The merchant's name, e.g. "B&H Photo". null when not resolved. |

An id that is unknown, expired, or could not be resolved answers `{ "resolved": false, "url": null, "retailer": null }`; it is never an error.

### Example

```
resolve_listing({ "resolve_id": "7c1f0e2b4d9a4f3e8b6a2c5d9e1f0a3b" })
```

```json
{
  "resolved": true,
  "url": "https://www.walmart.com/ip/Soundcore-Q20i/...",
  "retailer": "Walmart"
}
```

## find_cheaper_feedback

Records how a `find_cheaper` search went: which cards were useful, which do not belong in their list, a cheaper listing found elsewhere, what was missing. It is the Feedback endpoint as a tool, and the server's one write tool: it records a report with Findcheap and changes nothing else. A call costs nothing, and the first report on a search gives 1 credit back.

The tool names its search by `url`, the product page address the search was run on, because an assistant is shown no call id. The account's latest search of that address in the last 20 minutes is the one meant.

The limits are the endpoint's: one report per search, within 20 minutes of it; 1 credit back for the first report on a search, up to 100 credits a day per account, counted on the UTC day; and an account can switch feedback off in its dashboard, under Account.

The tool's description says what it records. It does not tell an assistant to send a report. To have an assistant send one after each search, give it the Feedback section of the agent skill, https://findcheap.ai/skills/findcheap/SKILL.md, or call the tool from your own code.

### Parameters

- `url` (string, required) - The product page address the search was run on: the `url` you gave `find_cheaper`.
- `rating` (`"good"` | `"partial"` | `"bad"`, required) - How the search went: `"good"`, `"partial"` or `"bad"`.
- `useful_cards` (object[]) - The cards you showed or recommended, each named by its list and its position in it. Each entry is an object with `list` (`"same_product"` or `"similar_products"`) and `position` (the card's place in that list, counting from 1).
- `wrong_cards` (object[]) - Cards that do not belong in their list, each named by its list and its position in it, with a reason. Each entry is an object with `list`, `position` and `reason`: `"different_product"` (it is not the product the page sells), `"not_equivalent"` (it is not an acceptable equivalent), `"price_differs"` (the listing's price is not the card's), `"unavailable"` (the listing is gone or out of stock) or `"other"`.
- `cheaper_elsewhere` (object[]) - Cheaper listings you found that the answer did not have, each with its address and its price. Each entry is an object with `url` (the cheaper listing's address) and `price` (its price, in the market's currency; null when you do not know it).
- `missing` (string[]) - What you expected and did not get, one short sentence each.
- `goal` (string) - What the shopper was trying to do, in one sentence. Nothing personal about the shopper.
- `note` (string) - Anything else worth knowing about this search, short.

Send the rating and at least one other field. Each list holds at most 20 entries. Text longer than its limit is cut, not refused: `goal` and each `missing` sentence at 300 characters, `note` at 1,000.

### Result

The Feedback endpoint's answer: `recorded`, `already_reported`, `credits_refunded`, `refunded_today`, `daily_refund_cap` and `daily_cap_reached`. Every field is documented under Feedback in the API reference.

### Description

What the tool tells an assistant about itself, word for word:

```
Record how a find_cheaper search went, so Findcheap can improve its results.

Give `url`, the product page address the search was run on, and `rating` (`good`, `partial` or `bad`), with at least one of: `useful_cards` (the cards you showed or recommended), `wrong_cards` (cards that do not belong in their list, each with a `reason`), `cheaper_elsewhere` (a cheaper listing you found that the answer did not have), `missing` (what you expected and did not get), `goal` (what the shopper was trying to do, in one sentence) and `note`. A card is named by its list (`same_product` or `similar_products`) and its position in that list, counting from 1. Keep it short: no page contents, and nothing personal about the shopper.

A report counts for a search made in the last 20 minutes, once per search; a second report on the same search changes nothing. The first report on a search gives 1 credit back, up to 100 credits a day. Answers `recorded`, `already_reported`, `credits_refunded` and `daily_cap_reached`.
```

### Example

```
find_cheaper_feedback({
  "url": "https://findcheap.ai/sample-product",
  "rating": "partial",
  "useful_cards": [{ "list": "same_product", "position": 1 }],
  "wrong_cards": [{ "list": "similar_products", "position": 1, "reason": "not_equivalent" }],
  "goal": "The same shoes for under 100 dollars."
})
```

```json
{
  "recorded": true,
  "already_reported": false,
  "credits_refunded": 1,
  "refunded_today": 1,
  "daily_refund_cap": 100,
  "daily_cap_reached": false
}
```

## The call's id and cost

Every result carries two values in its `_meta`, the part of a tool result that a client application can read and an assistant's model is not shown:

| Key | Type | Meaning |
| --- | --- | --- |
| `ai.findcheap/request_id` | string | The call's id (a UUID), the one on the account's usage ledger. Quote it when writing to support about the call. |
| `ai.findcheap/credits_used` | number | What the call cost the account: 10 for a search, 1 for a page read, 1 for a link that was resolved, 0 for one that was not, 0 for feedback. |
| `ai.findcheap/error` | string | On a tool error only: the error's type, e.g. `not_searchable` (see Errors below). |
| `ai.findcheap/searched_at` | string | On a `find_cheaper` answer only: when the search was made, in UTC (`2026-11-14T18:05:09Z`). The prices and links are those of that moment. |

A tool error carries the first two as well: the cost is 1 for `not_searchable` and `refused` (the page was read before the search stopped) and 0 for every other error, and the id is absent only when the call was refused before it began (`unauthorized`, `rate_limited`, `account_suspended`, `account_service_unavailable`). The dashboard (https://findcheap.ai/dashboard) lists every recorded call with its id.

## Errors

A tool call that cannot be served returns a tool error. Its text is one plain sentence the assistant can read out to the person it is helping, e.g. `This is not a product page, so there is nothing to search for. Send the address of an individual product page to search.` The sentence names no error type and no cost. The type is in the result's `_meta` under `ai.findcheap/error`. The types are the API's (see Errors in the API reference); the API's own messages are worded for developers and differ from these sentences:

| Type | Tool | When |
| --- | --- | --- |
| `invalid_url` | all | The url is not a web address we can fetch; the sentence says why. |
| `unknown_market` | all | The market is not one the agent serves. |
| `invalid_request` | all | A parameter has the wrong shape, e.g. an unknown marketplace name. |
| `not_searchable` | find_cheaper | The page is not a readable product page; the sentence says why. |
| `refused` | find_cheaper | A used-condition search for a grocery or consumable product, or for a product with no determinable brand. |
| `unreachable` | find_cheaper, read_product_page | The page could not be fetched right now; try again later. |
| `timeout` | find_cheaper, read_product_page | The page was reached but could not be read within 30 seconds, or the search did not finish within 60 seconds; try again. |
| `rate_limited` | all | The account is over one of its per-minute limits; the sentence says how long to wait. |
| `credits_exhausted` | all | The account has used its plan's credits for the period; the sentence names the plan and the reset date (monthly from the sign-up date on the free plan, the renewal on a paid one). |
| `unauthorized` | all | The key is missing or wrong (also a 401 when the client connects). |
| `account_suspended` | all | The account is suspended; the sentence says whom to write to. |
| `account_service_unavailable` | all | The key could not be checked right now; try again in a moment. |
| `search_not_found` | find_cheaper_feedback | The account made no search of that address in the last 20 minutes; no feedback is recorded. |
| `feedback_off` | find_cheaper_feedback | Feedback is switched off for the account. |

## Rate limits

Limits are per account and per minute (the plans are on the developers page, https://findcheap.ai/developers.md). An account has one budget: a `find_cheaper` call counts as a Search request, `read_product_page` as an Extract request, `resolve_listing` as a Resolve request and `find_cheaper_feedback` as a Feedback request, whether made through the MCP server or the API.

| Plan | Credits a month | Searches/min | Other requests/min |
| --- | --- | --- | --- |
| Free | 1,000 (100 searches) | 5 | 30 |
| Developer | 5,000 (500 searches) | 20 | 100 |
| Startup | 40,000 (4,000 searches) | 150 | 300 |
| Growth | 250,000 (25,000 searches) | 1,000 | 1,000 |
| Enterprise | Custom | Custom | Custom |
