What Is an MCP Server, and Why Does Your Business Website Need One?
Short version: llms.txt lets an AI agent read your site. An MCP server lets it act on it. Those are two different problems, and most of the current "AI-readiness" conversation only covers the first one.
The protocol, in plain terms
MCP stands for Model Context Protocol — an open standard, originally released by Anthropic, for how an AI assistant connects to external tools and data sources. Before MCP, every integration between an AI model and an outside system (a database, an API, a website's backend) was a custom, one-off job. MCP standardizes that connection: a server describes what it can do in a predictable format, and any MCP-compatible AI client — Claude, and increasingly others — can discover and call those capabilities without custom glue code for each one.
If that still sounds abstract: think of it as the difference between a brochure and a form. A brochure (your webpage, your llms.txt) tells a visitor — human or agent — what you do. A form lets them actually request something specific and get a structured answer back. An MCP server is how you give an AI agent a form instead of just a brochure.
What this looks like in practice
We run one ourselves, so this isn't theoretical. Perceptron Solutions exposes an MCP server at perceptron.solutions/mcp, described by a server card at /.well-known/mcp/server-card.json. It offers seven tools an AI client can call directly:
scan_site— runs our Agent-Ready Scanner against any URL and returns a Level 0–4 verdictlist_services— lists our consulting services and engagement optionsget_service_details— scope, ideal customer, and outputs for a specific servicerecommend_service— maps a described business problem to the most relevant serviceget_scan_remediation— explains how to fix a specific failed agent-readiness checkgenerate_readiness_report— a full scan with a polished, prioritized reportrequest_consultation— captures a consultation request
A person using an MCP-aware AI client can ask something like "check if this company's website is agent-ready and tell me what to fix" and the assistant can actually run scan_site and hand back a real result — not a guess based on what it can see on the rendered page.
Why this matters for an ordinary business, not just AI companies
As more people route research, comparisons, and even purchases through an AI assistant instead of a browser tab, a business that's readable-but-not-actionable to agents is leaving the second half of the interaction on the table. The agent can tell someone what you do. It can't check availability, get a quote, or start a request — because there's nowhere for it to actually do that. An MCP server is how you close that gap, the same way an API closed it for software-to-software integration a decade ago.
It's not a replacement for llms.txt or a well-structured site — it's the next layer on top of them. Read → understand → act. Most sites today stop at "read," if they get that far.
Want to see where your own site stands before building anything new? Our free scanner checks agent-readiness — including whether you have anything like this at all — and gives specific, prioritized fixes.
Run the Agent-Ready Scanner →We help teams design and build these — from a first llms.txt to a full MCP server exposing real business actions. If that's useful, get in touch.