Most of the conversation about MCP focuses on consuming: connecting an agent to tools and data that already exist, such as an ERP or CRM. But there is the other, less discussed side: a company can also expose its own MCP server, making its data and processes queryable by any compatible AI agent — whether an internal agent, a partner's, or even a customer's.

What "having" an MCP server means

An MCP server is, in practice, a standardized door the company opens so AI agents can ask structured questions and receive structured answers — instead of an agent having to "read" the site or guess information from loose text.

A concrete, real example: Yan.m maintains its own public MCP server at yanmantovani.com/api/mcp, using Streamable HTTP transport and JSON-RPC 2.0. It exposes tools such as get_agency_info, list_services, list_products, list_case_studies, and get_contact. That means any MCP-compatible agent — from a client evaluating vendors, a partner doing integration, or an automated research tool — can directly query which services the agency offers, what the ecosystem products are, or how to get in touch, without scraping the site or depending on the content being well indexed in a traditional search engine.

Why this is different from just having a well-made site

A site is built to be read by humans browsing visually. An MCP server is built to be queried by agents, in a structured and predictable way. They are not substitutes for each other — they are different layers of the same digital presence. As more AI tools act as intermediaries between a person and the information they seek, having structured data accessible via a standardized protocol stops being a technical detail and becomes a way to ensure the right information arrives without distortion.

Who should consider this

It makes sense to evaluate your own MCP server when the company:

  • Has structured processes or catalogs that make sense to query programmatically (services, products, availability, order status, technical documentation).
  • Already works or plans to work with partners, integrators, or clients who use AI agents to automate part of their own work.
  • Wants to position itself technically and visibly for the AI ecosystem forming, not only for traditional search engines.

Who does not need to worry about this yet

On the other hand, it does not make sense to invest in this now if the company:

  • Does not have enough structured processes or data to justify an API — in that case, the priority is organizing information internally before exposing anything.
  • Has a small enough operation that a well-made site and direct service (WhatsApp, for example) already solve the information access problem.
  • Is still validating its own business model — exposing infrastructure before stabilizing what it represents is usually effort out of order.

The balance point

Having an MCP server is not about following trends; it is about reducing friction for whoever — human or agent — needs reliable information about the company. That kind of architecture decision, evaluating what is worth structuring now and what can wait, is part of Yan.m's MCP & Integrations work with each client, case by case.

For those who want to understand the protocol fundamentals before deciding, it is worth starting with the official Model Context Protocol documentation and Anthropic's material on the topic.