A quiet but significant shift is under way in how purchasing decisions get made online. Increasingly, consumers are delegating research, comparison, and even transaction completion to AI agents — software that acts autonomously on their behalf. This is not a distant prospect. With Anthropic's Model Context Protocol (MCP) gaining rapid adoption as a de facto standard for how AI agents interact with external tools and services, the infrastructure for agentic commerce is being assembled right now. The organisations that prepare their digital estate for this traffic early will enjoy a structural advantage that latecomers will find difficult to close.
Think of it as a new category of visitor arriving at your site — one that does not browse, does not respond to visual merchandising, and has no patience for unclear product data or broken API responses. It reads structured information, evaluates it programmatically, and moves on in milliseconds. If your site cannot communicate clearly with that visitor, it will simply be skipped in favour of a competitor whose infrastructure can.
Understanding MCP and Why It Changes Ecommerce
The Model Context Protocol is an open standard that defines how AI models connect to external data sources and tools. Rather than scraping web pages or relying on ad-hoc integrations, an MCP-compatible agent can call a structured endpoint, receive a well-formed response, and take meaningful action — whether that means retrieving product availability, comparing specifications, or initiating a purchase flow. Anthropic introduced MCP, but the protocol is being adopted across the industry with notable speed, with integrations appearing in developer tooling, enterprise platforms, and consumer AI assistants.
For ecommerce operators, this matters because MCP shifts the locus of product discovery. Historically, winning in product search meant optimising for human eyes on a search engine results page. In an agentic model, an AI assistant might query multiple MCP-compatible retailers simultaneously, evaluate responses against the user's stated preferences, and recommend or complete a purchase — all without a human loading a single product page. Your visibility in that process depends entirely on whether your site exposes the right endpoints in the right format.
The Role of llms.txt in Agentic Discovery
Alongside MCP endpoints, a simpler but equally important convention is emerging: the llms.txt file. Analogous to robots.txt in function but aimed at large language models rather than crawlers, llms.txt allows a site to declare, in plain language, what it offers, how its content is structured, and what an AI agent should know before interacting with it. The format is lightweight — a markdown-style document placed at the root of your domain — but its strategic value is substantial.
An llms.txt file that clearly describes your product catalogue structure, your API capabilities, your returns policy, and your checkout mechanism gives an AI agent the context it needs to prefer your site over a less legible competitor. Early adopters in sectors such as consumer electronics, apparel, and speciality retail are already experimenting with this. For UK organisations operating in competitive categories, publishing a well-considered llms.txt file is a low-effort, high-leverage action that can be taken independently of broader platform work.
Auditing Your Product Data and API Architecture
The readiness of your product data is the foundation everything else builds on. AI agents cannot compensate for incomplete or inconsistent structured data — they will simply deprioritise results they cannot confidently interpret. A practical audit should examine several dimensions: attribute completeness (do all products have accurate specifications, dimensions, materials, and compatibility data?), schema markup (are you using Schema.org Product markup correctly and comprehensively?), API reliability (are your product and inventory endpoints stable, authenticated appropriately, and returning consistent response formats?), and pricing accuracy (is real-time stock and pricing information accessible without requiring a user session?).
Many ecommerce platforms in use across UK retail — Shopify, Magento, commercetools, and bespoke builds alike — can expose MCP-compatible endpoints either natively or through middleware. The key architectural question is whether your API layer was designed with machine consumption in mind or retrofitted from a UI-first codebase. The latter often produces inconsistencies that human shoppers overlook but agents will flag as unreliable. A focused technical review now, before agentic traffic is measurable in your analytics, is considerably less disruptive than a reactive overhaul later.
Trust, Authentication, and the Agent Relationship
Agentic commerce raises questions that go beyond data structure. When an AI agent initiates a transaction on a user's behalf, questions of trust, consent, and authentication become commercially and legally material. How does your platform verify that an agent is acting with genuine user authorisation? How do you handle returns or disputes initiated by an agent rather than a human? These are not purely technical questions — they touch on your terms of service, your fraud prevention logic, and potentially your obligations under UK consumer protection regulation.
Forward-thinking teams are beginning to define agent-specific API scopes — permissions that allow read access to product data and inventory without granting unconstrained transactional authority, with additional authentication steps required before a purchase is confirmed. This mirrors the principle of least privilege familiar from enterprise security practice. Building this architecture now means you can participate in agentic commerce on your own terms, rather than scrambling to retrofit controls after an incident.
The window to act ahead of the curve here is real but not indefinitely open. MCP adoption is accelerating, consumer AI assistants with shopping capabilities are being released by major platform providers, and the sites that have invested in structured, machine-readable product data will compound that advantage with every passing month. The practical starting point for most organisations is a two-track effort: publish an llms.txt file that accurately describes your catalogue and capabilities, and commission a structured audit of your product data quality and API architecture against MCP compatibility requirements.
At iCentric, we work with ecommerce and digital product teams across the UK to assess readiness for emerging AI-driven traffic patterns and to build the API and data infrastructure that supports them. If you are considering where agentic commerce fits in your digital roadmap, we are glad to have that conversation.
What is the Model Context Protocol (MCP) in simple terms?
MCP is an open standard that defines how AI agents communicate with external services and data sources. Instead of scraping web pages, an MCP-compatible agent calls a structured API endpoint and receives a machine-readable response it can act on. Anthropic created it, but it is being adopted broadly across the AI industry.
How soon is agentic shopping traffic likely to affect UK ecommerce sites?
Agentic shopping is already live in limited form through AI assistants from major platform providers. Measurable traffic impact at scale is likely within one to three years for most categories, but competitive advantage accrues to early movers. Organisations in high-consideration categories such as electronics, home goods, and B2B supplies are likely to see the effect sooner.
Do I need to build a custom MCP server, or can existing platforms support it?
Many established ecommerce platforms — including Shopify, commercetools, and Magento — can expose MCP-compatible endpoints through plugins, middleware, or custom API layers without a full rebuild. The extent of custom work depends on how cleanly your current API architecture separates data from presentation logic.
Is llms.txt an official standard, and will AI companies actually use it?
The llms.txt convention is an emerging community standard rather than a ratified specification, similar to how robots.txt began. Several AI companies and tool developers have indicated support for it, and its adoption is growing. Even if uptake varies across AI providers, publishing a well-structured llms.txt file carries minimal cost and tangible upside.
How does agentic commerce affect SEO and traditional product discovery?
Agentic commerce adds a layer of discovery that sits alongside, rather than replacing, traditional search. Classic SEO remains important for human-initiated searches, but product visibility in agent-mediated purchasing depends on structured data quality and API accessibility rather than keyword ranking. Both disciplines need to be maintained in parallel.
What consumer protection or legal considerations apply when an AI agent completes a purchase?
Under UK consumer protection law, the contractual relationship is generally between the retailer and the human consumer, regardless of whether an agent initiated the transaction. However, operators should review their terms of service to ensure they clearly address agent-initiated purchases, and should consider how their dispute and returns processes handle cases where no human directly participated in the checkout flow.
How should we handle pricing and inventory data for agentic traffic?
Agents expect real-time, session-independent pricing and stock data. If your pricing or inventory requires a logged-in session or is only rendered client-side via JavaScript, agents will likely receive inaccurate or incomplete information. Exposing a server-side API endpoint that returns current pricing and availability without authentication prerequisites is the recommended approach.
Can agentic commerce benefit B2B ecommerce as well as B2C?
B2B is arguably a stronger early use case for agentic purchasing. Procurement agents can be tasked with finding compliant suppliers, comparing specifications against requirements, and initiating purchase orders — all without manual effort. B2B operators with complex catalogues, contract pricing, and approval workflows stand to gain significantly from MCP-compatible infrastructure.
What metrics should we track to monitor emerging agentic traffic?
Standard web analytics will not reliably distinguish agent traffic from bot traffic initially. Organisations should instrument their API endpoints separately, logging user-agent strings, request patterns, and session characteristics to identify non-human access. Over time, AI platform providers are likely to introduce identifiable agent signatures, similar to how Googlebot is identifiable today.
Should smaller UK ecommerce operators prioritise this now, or is it only relevant to large retailers?
Smaller operators in niche or specialist categories may actually benefit disproportionately, because agents searching for specific products within a narrow category are more likely to surface a well-structured specialist retailer over a large generalist. The investment required to publish llms.txt and improve product data quality is modest and within reach for businesses of most sizes.
Get in touch today
Book a call at a time to suit you, or fill out our enquiry form or get in touch using the contact details below