MCP: Model Context Protocol — Let AI Agents Take Action
TL;DR
MCP (Model Context Protocol) lets AI agents not just read your site but interact with it — book appointments, make purchases, access data. It is the action layer of the agentic web.
Last updated: 2026-03-09
What Is MCP?#
MCP stands for Model Context Protocol. It is an open standard developed by Anthropic that defines how AI agents connect to external tools, data sources, and services. If llms.txt tells AI what your site is about and NLWeb lets AI ask questions, MCP lets AI take action.
With MCP, an AI assistant can do more than describe your restaurant. It can check table availability, make a reservation, and send a confirmation. It can browse your product catalog, add items to a cart, and process a checkout. It turns your website from a passive source of information into an active participant in AI-driven workflows.
MCP uses a client-server architecture. The AI model is the client, and your website or service acts as the server. The server exposes a set of tools (functions the AI can call), resources (data the AI can read), and prompts (templates the AI can use).
How MCP Works: Tools, Resources, and Prompts#
MCP defines three core concepts that structure how AI agents interact with your service.
Tools are functions the AI can call. For example, a hotel might expose tools like
check_availability, book_room, and cancel_booking. Each tool has a defined input schema (what parameters it accepts) and output schema (what it returns).
Resources are data the AI can read. These are like read-only endpoints. A restaurant might expose its menu, hours, and location as resources. The AI can access this information without modifying anything.
Prompts are reusable templates that guide how the AI uses your tools and resources. They help ensure the AI asks the right questions and follows the right workflow when interacting with your service.- Tools — Actions the AI can perform (book, purchase, cancel, search)
- Resources — Data the AI can read (menus, catalogs, schedules, FAQs)
- Prompts — Templates that guide AI behavior when using your service
Reading vs Acting: Where MCP Fits#
The three AI protocols form a stack, from passive to active.
llms.txt is the reading layer. It gives AI a static summary of your site. There is no interaction. The AI reads your file and uses it as context.
NLWeb is the question layer. It lets AI ask your site specific questions and get structured answers. There is interaction, but it is read-only.
MCP is the action layer. It lets AI perform tasks on your behalf. The AI can read data, call functions, and trigger real-world outcomes like bookings, purchases, or account changes.
Most sites should start with llms.txt (easy, high impact), then add NLWeb (moderate effort, valuable for content-heavy sites), and finally consider MCP (higher effort, essential for transactional sites).Use Cases#
MCP is most valuable for businesses where users take action through their website. Here are some real-world examples of how MCP changes the game.
- E-commerce: AI agents can browse products, check inventory, compare options, and complete purchases on behalf of users.
- Restaurants and hospitality: AI can check availability, make reservations, modify bookings, and handle cancellations.
- SaaS platforms: AI can query dashboards, pull reports, create tickets, and manage user settings through exposed tools.
- Healthcare: AI can check appointment availability, book visits, and retrieve patient-facing information (within compliance boundaries).
- Financial services: AI can retrieve account summaries, initiate transfers, and answer policy questions using structured resources.
- Real estate: AI can search listings, schedule viewings, and submit inquiries directly through your platform.
Current Adoption#
MCP was released by Anthropic in late 2024 and has seen strong momentum. Major companies including Block, Apollo, Replit, and Sourcegraph adopted MCP early. The protocol is open source and language-agnostic, with SDKs available for Python, TypeScript, Java, and other languages.
As of early 2026, MCP is supported natively by Claude (Anthropic’s AI), and several other AI platforms have announced support or compatibility. The ecosystem of MCP servers is growing rapidly, with community-built integrations for databases, APIs, CMS platforms, and more.
The AgentReady™ scanner evaluates MCP readiness as part of the AI Protocols factor. While MCP is not yet essential for every site, businesses with transactional functionality should evaluate it seriously.
Related Pages
Frequently Asked Questions
Do I need MCP for a blog or content site?
Probably not right now. MCP is most valuable for transactional sites where users take action (buy, book, subscribe). For content sites, llms.txt and NLWeb provide more immediate value.
Is MCP secure?
MCP includes authentication and authorization mechanisms. You control which tools are exposed and what permissions AI agents have. Always follow security best practices when exposing any API.
Can MCP work with my existing API?
Yes. MCP servers often wrap existing APIs. If you already have a REST or GraphQL API, you can create an MCP server that exposes your existing endpoints as tools and resources.
Was this page helpful?