For founders, technical marketing leads, and developer-adjacent business owners. Forward-looking, already AI-aware, wants to understand the infrastructure layer before it becomes obvious — and wants an honest answer to whether this is right for their business.
The question everyone is asking — and what it actually means for your business
If you have been on LinkedIn recently, you have seen MCP everywhere. Model Context Protocol. Everyone has an opinion. Half of them are wrong, and almost none of them answer the questions that actually matter to you as a business owner: how does this relate to SEO and the AI visibility work I’m already doing — and should my business actually pursue this, or is it a distraction dressed up as strategy?
For the last twenty years, online visibility meant ranking in search engines. Google, Bing, position one on a results page. Traditional SEO.
For the last two to three years, a new layer has emerged on top of that. AI systems are now answering questions directly — through Google’s AI Overviews (AIO), through Perplexity and ChatGPT (AEO and GEO), through AI agents operating autonomously (AAO). Being visible in these answers requires different signals than traditional SEO: entity clarity, structured content, external corroboration, consistent identity across the web. This is the work that determines whether an AI recommends your business when someone asks for a shortlist. It builds on SEO. It does not replace it.
MCP is the third layer. AI systems are moving from answering questions to completing tasks — from ‘here are three options’ to ‘I’ve checked availability and here are two that can deliver by Thursday.’ MCP is the infrastructure that makes your business transactable — not just findable, not just recommendable, but a business an AI agent can actually act on behalf of. Whether that is what your business needs is a different question entirely.
The progression is: rank in search → appear in AI answers → be executable by AI agents. MCP is stage three. You cannot skip to it. And whether stage three belongs in your strategy depends on what it would actually do for your conversion model.
What MCP actually is — in plain terms
Think of MCP as a USB-C port for AI. Just as USB-C created a universal standard that allowed any device to connect to any cable without a custom adapter, MCP creates a universal standard that allows any AI system to connect to any business’s live data without a custom integration built from scratch.
Before MCP, connecting an AI system to business data required a bespoke API for each AI platform — time-consuming, expensive, and different for every tool. With MCP, you build one server — one ‘toolbox’ the AI can open — and any MCP-compatible AI can plug directly in. The result is an AI that doesn’t just know about your business from what it read six months ago. It knows what is true right now: today’s pricing, current availability, up-to-date credentials.
Major platforms are building toward MCP. Google is adding support in Gemini for enterprise users. Microsoft has introduced it in Copilot Studio. Amazon Ads launched an MCP server in 2026 for AI-managed campaign strategies. OpenAI formally adopted MCP in March 2025 across its Agents SDK and desktop app. The standard is moving from experimental to mainstream — though implementation depth and maturity varies considerably by platform. (Perplexity, notably, announced in March 2026 that they are stepping back from MCP toward a traditional API, citing token consumption as the primary concern — a signal worth watching.)
Why this changes the competitive landscape
The temptation is to frame MCP as an accuracy tool. That benefit is real — when an AI is grounded in your actual database, it is less likely to misrepresent your products or pricing. But the more important implication is competitive executability. AI systems recommend what they can justify. Live, current, structured data is easier to justify than historical inference.
Consider the difference directly. A buyer asks an AI assistant to help procure office supplies within a budget. The AI finds three suppliers. One has an MCP server — the AI can verify stock, check pricing, and complete the order in the same conversation. The other two require the buyer to leave the AI interface and do it manually. Which supplier gets the transaction?
The closed loop — and why the position race has already started
MCP creates a closed loop: visibility → action → reinforcement. The more transactions an AI completes with a business through MCP, the more confident it becomes recommending that business for similar queries. The signal reinforces itself with every interaction.
Businesses inside that loop get stronger with every transaction. Businesses outside it get compared — but never chosen.
This is not a feature race. It is a position race. By the time MCP feels obvious, the advantage will already be allocated. AI systems don’t just evaluate options — they learn from interaction. The businesses connected first become the ones the system knows how to use. And systems prefer what they already know works.
This isn’t theoretical. When a buyer asks an AI to shortlist suppliers, the system won’t just ask ‘who is relevant?’ — it will ask ‘who can I actually complete this with?’ If your business isn’t connected, you’re not excluded. You’re bypassed.
Where MCP sits in the visibility stack — and why you cannot skip to it
MCP is Stage 5 of the AI recommendation pipeline. It is the live data and executability layer. But stages one through four have to be solid first:
Stage 1 — Recognition: Consistent entity identity across your website, Google Business Profile, industry directories, and regulatory listings. Without this, the AI cannot safely name you.
Stage 2 — Verification: Independent third-party evidence — press coverage, directories, industry body membership, case studies. The AI is evaluating whether it is safe to name you. Third-party corroboration is the only currency that buys that safety.
Stage 3 — Selection: Specific enough positioning to match a precise query. Generic descriptions give the AI nothing to match against. Explicit, consistent positioning in the language clients use creates eligibility.
Stage 4 — Citation: Content structured so a machine can extract it. FAQ sections, schema markup, direct answers. Flowing prose that buries key facts fails machines.
Stage 5 — MCP: Live, queryable data. The infrastructure that makes you transactable rather than just recommendable.
MCP doesn’t make you recommendable. It makes you executable. If an AI doesn’t already trust and understand your business, MCP simply gives it faster access to something it won’t choose.
Stages one through four determine whether AI systems name you at all — in AI Overviews, Perplexity answers, and autonomous agent recommendations. Stage five determines whether, once named, the AI can act. Fix the foundations. Build the structured content. Establish the entity signals. Then build toward the live layer.
For the practical companion — LLM landscape table, 10-point readiness checklist, entry points by business type, and governance template — see MCP Readiness: Where to Start.
Who actually benefits — and who doesn’t
Category One: High MCP benefit — standardised, transactable, self-serve
Businesses where an AI agent can complete a meaningful part of the transaction: verify specs, check availability, compare pricing, place an order, book an appointment. E-commerce, self-serve SaaS, booking-based services, and commoditised B2B procurement. Shopify stores already have a native MCP endpoint enabled by default — this is not a future consideration for this category, it is a present one.
Category Two: Partial MCP benefit — high-consideration, human conversion
Businesses where the research phase is long and AI shortlisting matters, but the conversion is human. Premium bespoke services, complex B2B SaaS with assisted sales, high-consideration B2C. MCP’s role here is shortlist improvement, not transaction completion. Having queryable, current data about credentials, accreditations, and availability makes a business more credible and actionable than a competitor relying only on what AI scraped months ago.
Where MCP would hurt this category: if it enabled a self-serve interaction that bypasses the human relationship driving conversion. For businesses with an assisted sales model, an MCP implementation that made the business look self-serve would damage the commercial model it is designed to support. The governance question is not ‘should we use MCP’ but ‘exactly which data layer is safe to expose, and which must stay behind a human conversation’.
Category Three: Low or wrong MCP benefit — regulated, relationship-driven, high-trust
Businesses where AI intermediation in the conversion process damages the value proposition — because it creates regulatory risk, or because the trust signal that drives conversion is specifically that a qualified human is involved.
Law firms and regulated legal services are the clearest example. The conversion mechanism for legal representation is a qualified, regulated solicitor making contact. An AI agent that ‘completed a transaction’ for legal representation would potentially provide regulated legal advice without qualification, breach conduct rules, and destroy the trust signal that makes the firm worth calling in the first place. The reputational cost of getting this wrong is disproportionate to any shortlisting benefit.
Where a strictly limited MCP implementation could be appropriate in regulated services: structured exposure of practice areas, team credentials, fee transparency where regulations permit, and appointment booking for an initial consultation. The line is clear: factual information about the firm, not professional services delivered through the protocol.
The business case — stress-tested
Strengths
First-mover compounding: early adopters accumulate interaction history and system trust that is difficult to displace once established. Closed-loop reinforcement: every successful AI-assisted transaction increases the likelihood of future recommendations. Data structure investment pays dividends regardless — well-structured data improves recommendation eligibility across all stages. Reduces hallucination risk. Competitive gap: very few SMBs have implemented MCP in 2026.
Weaknesses
Implementation cost is real: a custom MCP server requires developer time (£5,000–£25,000+ depending on complexity) and ongoing governance overhead. Standard immaturity: MCP is evolving rapidly and implementations may require rework. Platform tooling for configuration-level MCP does not yet exist for most non-Shopify businesses. Benefit depends entirely on stages 1–4 being solid first. ROI measurement is currently very difficult.
Opportunities
Platform adoption window (2027–28): businesses that understand MCP now will implement faster when configuration-level tooling arrives. Competitive gap while most businesses sleep: the majority will not act until MCP feels obvious — by which point the advantage is allocated. Data structuring work compounds across every layer of the visibility stack simultaneously.
Threats
Security breach from poor governance: MCP gives AI agents a direct connection into company systems — insufficient permission scoping creates a genuine attack surface. Regulatory exposure in professional services. Reputational damage if an MCP-connected AI misrepresents pricing or capabilities to a customer. Standard fragmentation risk if MCP does not achieve universal adoption. Prompt injection attacks through MCP-connected systems are an active security concern, not theoretical.
The questions a CFO would ask
What specific transaction are we trying to enable? Vague answers are not sufficient. If you cannot state the specific use case, MCP is not the right next investment.
What do stages 1–4 look like right now? Run the AI Recommendation Readiness Diagnostic. If there are gaps in entity consistency, corroboration, or structured content — close those first.
What is the governance position? Before any code: what can the AI read, write, trigger, and what is off limits? Who owns the review cadence? What is the incident response?
What does wrong look like? Map the failure modes. For each: what is the reputational, regulatory, and customer consequence?
What is the cost of waiting versus acting poorly? For most businesses in 2026, acting poorly costs more than waiting and building foundations first.
The honest conclusion
MCP is not the most urgent thing on every business’s list in 2026. For some businesses, it is genuinely not the right investment at any point in the near term. For others, it is the most significant competitive infrastructure shift of the decade. The businesses that will use this window well are the ones who answer honestly: what does my conversion model require? What would being transactable by an AI agent mean for my business — and is that a benefit, a risk, or something that conflicts with why clients choose me?
MCP doesn’t change whether your business exists. It changes whether your business can be used. In a system that increasingly values execution over explanation, that difference is what determines who gets chosen — and who gets left behind.
This article is part of the How to Make LLMs Recommend Your Business content programme. For the practical companion — LLM landscape table, readiness checklist, entry points by business type, and governance template — see MCP Readiness: Where to Start. Take the AI Recommendation Readiness Diagnostic to find your primary bottleneck.
