Particular Audience Particular Audience
Talk to us

AI for search, personalisation and retail media

A different store for every customer.

Bring what each shopper wants together with what your business needs.

Search, recommendations, retail media and promotions across your digital store and AI shopping experiences.

Adaptive Transformer Search MCP — Product Search for AI Agents | Particular Audience

An MCP server that gives AI assistants and agents Adaptive Transformer Search: natural-language product search, filters, sorting, merchandising control and relevant sponsored results.

MCP server

Give AI agents your search — with your relevance and your merchandising rules.

The Adaptive Transformer Search MCP server lets an AI assistant or agent search your catalogue through Particular Audience's Adaptive Transformer Search. It understands what the shopper means, not just the words, and keeps your boosts and sponsored results in the answer.

Status: Open source server on GitHub; connects to your PA Search API.

Tools: Search — Natural-language search with optional filters and paging. Filtered search — Search with required filters, still understanding the meaning of the query. Sorted search — Search with custom sorting such as price or popularity.

What comes with it: Meaning, not keywords — Knows "chocolate milk" from "milk chocolate" and handles misspellings and plain-language requests. Merchandising control — Boost products, brands or categories for promotions, stock or margin — the agent sees your ranking. Sponsored results by relevance — Sponsored products appear only when they match what was asked. Handled for you — Authentication, error handling and retries are built into the server.

Today: Agent reads a static feed, Gets list prices and keywords, Misses your promotions, Picks its own ranking, Your rules don't apply. With MCP: Agent calls your search, Gets your ranking, merchandising and relevant sponsored results.

Shopper agent: “A backpack that fits a 16-inch laptop.” The assistant gathers: Meaning-aware results from your catalogue, with filters and sorting. What it unlocks: AI shoppers find the right product from you — not a guess from a feed.

Retailer app: “Gluten-free pasta, cheapest first.” The assistant gathers: Filtered, sorted search with your relevance. What it unlocks: Plug your search into your own assistant or app with ready-made tools.

Your merchandising rules: “Running shoes” — during your spring promotion. The assistant gathers: Results ranked with the boosts you set for promotions, stock or margin. What it unlocks: Agents see the same ranking shoppers see on your site.

Sponsored results: “Dishwasher tablets” — with a relevant sponsored product. The assistant gathers: Sponsored products that match the request, labelled. What it unlocks: Sponsored results in agent answers — only when relevant.

What it unlocks: Be found by AI shoppers — Agents search your catalogue with meaning, not just keywords. Keep your ranking — Boosts and merchandising rules travel with every agent search. Monetise agent answers — Relevant sponsored results inside the conversation. Ready-made tools — Search, filtered search and sorted search out of the box. Open source — Read the code, run it your way. Foundation for agentic commerce — The base for Functional Ad Units, Sponsored Prompts and price match for agents.

Setup: 1. Install from GitHub — Get the open-source server. 2. Add your Search API keys — Credentials stay on your side. 3. Agents search — Your ranking and rules apply as on your site.

Trust: Credentials never sent to the assistant; Your relevance and merchandising rules; Sponsored only when relevant; Open-source code.

Who it's for: Retailers — Put your own search in front of AI shoppers and agents. Product & engineering teams — Ready-made search tools for your own assistant or app. Merchandisers — Your boosts for promotions, stock or margin, kept in agent answers. Retail media teams — Relevant sponsored results inside agent conversations.

What's in the answer: Products ranked by meaning, not just keywords; Your boosts and merchandising rules applied; Sponsored products only when they match the request; Filters, sorting and paging.

What it doesn't do: Changing your prices; Inventing stock or products that aren't in your catalogue; Pushing sponsored items that don't match the request.

Works with any assistant or agent that supports MCP, including your own.

Getting started: Get Particular Audience Search API access → Install the open-source server from GitHub → Add your Search API credentials on your side → Connect your assistant or agent and search.

For IT and security: Open source — The server code is public on GitHub for your team to review. Credentials — Search API credentials stay on your side and are never sent to the assistant. Built in — Authentication, error handling and retries are handled by the server. Your rules — Ranking, boosts and sponsored rules come from your Particular Audience search setup.

Plain-language search: “Something warm for a winter hike” “A backpack that fits a 16-inch laptop”

Filtered: “Gluten-free pasta” “Running shoes in size 10”

Sorted: “Cheapest first” “Most popular dishwasher tablets”

How access works: Install the server from GitHub and connect it to your Particular Audience Search API credentials. Agents call the search tools; your ranking, boosts and sponsored rules apply as on your site. Keep the credentials on your side — they are never sent to the assistant.

Ask things like: “Comfortable walking shoes under $100” “Gluten-free pasta, cheapest first” “A backpack that fits a 16-inch laptop”

Source code: https://github.com/Particular-Audience/adaptive-transformer-search-mcp

Is it open source? Yes, the server code is public on GitHub. It needs Particular Audience Search API access to return your catalogue.

Do sponsored results take over? No. Sponsored products appear only when they genuinely match the request.

Can I keep my merchandising rules? Yes. Boosts for promotions, stock or margin apply to agent searches as they do on your site.

Which assistants or agents can use it? Any assistant or agent that supports MCP, including one you build yourself.

Do I need Particular Audience search? Yes. The server connects to the Particular Audience Search API to search your catalogue.

Where are the API credentials kept? On your side. They are never sent to the assistant.

Does it understand misspellings and plain language? Yes. It searches by meaning, so “chocolate milk” isn't confused with “milk chocolate”.

Can I use it in my own app? Yes. Connect your own assistant or app to the same search tools.

How does this relate to Functional Ad Units and Sponsored Prompts? This server gives agents your search. Functional Ad Units and Sponsored Prompts build on that to put relevant sponsored experiences inside AI conversations.

Does it change prices? No. It returns your catalogue as your search ranks it.

Adaptive Transformer Search MCP

Your search relevance and merchandising for AI agents.

Works with

Outcomes it moves