Asking an AI Assistant to Search Products Through MCP: Sample Prompts and What Comes Back
MCP, short for Model Context Protocol, is a way for an AI assistant to use outside tools. Once the Affiliate.com MCP server is connected, an assistant such as Claude or ChatGPT can search products, look up merchants and networks, and convert product identifiers while you chat. You describe what you want in plain English. The assistant turns that into a search, runs it against live data, and shows you the results.
This article is for people who would rather type a request than write code. It walks through sample prompts, what each one returned when we tested it, and a few habits that make the answers more useful.
Before you start
The server address is https://mcp.affiliate.com/all, and sign-in uses OAuth, so there is no API key to copy. An MCP subscription is required, and accounts without one receive a 403 error. The step-by-step connection for Claude is covered in How to Connect Claude to Affiliate.com’s MCP for AI Assisted Product Curation. If you are deciding between MCP and the REST API, MCP or REST API? compares the two.
Five kinds of questions you can ask
Each kind of question maps to a tool, and each tool uses a set number of credits. Checking your usage is free.
| What you want | Example prompt | Credits per call |
|---|---|---|
| Find products | Find new, in-stock wireless headphones under $150 that are on sale. | 4 |
| Compare brands | Compare the lowest in-stock prices for two running shoe brands. | 4 per search |
| Explore merchants | Which merchants with "outdoor" in the name do you carry, and how many products does each have? | 2 for a list, 1 per merchant detail |
| Convert identifiers | Turn these product page URLs into barcodes. | 1 |
| Check usage | How many credits have I used this month? | 0 |
Example 1: a deal search
The prompt we used:
Find new, in-stock wireless headphones from Sony, Bose, Sennheiser, JBL or Anker, priced in US dollars between $40 and $150, that are discounted 15% to 60%. Sort by biggest discount, remove duplicates, and show the retailer and network.
The search matched 349 products, and the assistant showed the top results:
| Product | List price | Sale price | Discount | Retailer | Network |
|---|---|---|---|---|---|
| Sennheiser CX 400BT true wireless earbuds | $199.95 | $86.99 | 56% | Big-box retailer | Impact US |
| JBL Endurance Peak 3 true wireless earbuds | $109.95 | $49.99 | 55% | Big-box retailer | Impact US |
| JBL SoundGear Sense open-ear headphones | $164.95 | $74.99 | 55% | Electronics retailer | Impact US |
| Sennheiser ACCENTUM true wireless earbuds | $219.95 | $99.95 | 55% | Department store | Impact US |
Two details are worth knowing. The same JBL headphones appeared a second time on a different network under a slightly different product name, so a duplicate can slip through. Asking the assistant to also compare product names catches those. We also ran the prompt once without the price and discount limits. The top result was a product with a $2,000 list price shown as 96% off, and several results came from European merchants. Setting a currency and capping the discount keeps results believable. The reasoning behind stacking conditions like these is explained in How to Layer Product Search API Filters for More Precise AI Shopping Results.
Example 2: comparing two brands
The prompt:
Compare the lowest in-stock prices for new Brooks and ASICS running shoes in US dollars.
Our first attempt sorted by lowest price and returned $1.99 heel inserts, accessories whose titles mention running shoes and whose seller name resembled the brand. Adding a $60 minimum price and the word "new" produced cleaner results:
| Brand | Lowest result | Price | What to notice |
|---|---|---|---|
| Brooks | Revel 8 running shoe, men’s | $64.98 | Each size is its own listing, so the same price repeats. |
| ASICS | Comfy Baby running shoes | $60.84 | These are toddler shoes. Adding "adult" or "men’s" to the prompt avoids them. |
The lesson is that a "cheapest first" sort rewards the lowest number, not the best match. A price floor and a clear description of the product type usually fixes it.
Example 3: exploring merchants
Asking "Which merchants with 'outdoor' in the name do you carry?" returned 71 matches. The first page included a sporting goods retailer with about 436,000 products on CJ, an outdoor retailer with about 11,700 products on AvantLink, and several smaller merchants with a few hundred products each. Counts like these were taken at the time of testing and change as feeds update. This kind of question is useful before a product search, because it shows which merchants have enough depth to be worth filtering on.
Habits that improve results
- Name the brand, price range and currency. Specific requests return specific results.
- Say "new" when you mean new. Refurbished listings can otherwise appear.
- Ask for a small number first. Five results are easier to check than fifty.
- Ask the assistant to show its filters. Seeing the exact conditions it used makes it easy to adjust them.
- Test in the Query Builder first. The Affiliate.com Query Builder shows the same filters visually, which helps when a prompt keeps returning the wrong thing.
What the assistant will not do
The search matches keywords and fields. It does not rewrite your question on its own, run semantic search, or search by image. The assistant does the translating from your words into filters, so a vague request produces a vague filter. A request that names the product type, brand, price and condition gives it the most to work with.
A prompt template to reuse
Use the Affiliate.com MCP to find [product type] from [brands or merchants].
Only show new items that are in stock, priced in [currency] between [low] and [high].
Limit discounts to [range] and remove duplicate products.
Sort by [price or discount] and show [number] results with the retailer, network and price.
List the filters you used.Fill in the brackets, paste it into a chat with the connector turned on, and adjust one condition at a time. Checking your credits afterward costs nothing, so it is easy to see what each style of question uses.