The Model Context Protocol (MCP) is an open protocol that lets AI assistants work safely with external systems: databases, CRMs, ERPs, issue trackers, online stores. In a year it went from experiment to industry standard: MCP is supported by Claude, ChatGPT, Gemini and Microsoft Copilot, and there are more than 10,000 public MCP servers.
In December 2025 the Linux Foundation announced the Agentic AI Foundation, which brings together MCP, goose and AGENTS.md, co-founded by Anthropic, Block and OpenAI. The protocol is now under neutral governance, a signal to businesses that it is safe to build on.
How it works
- An MCP server is a small application next to your system. It describes which tools and resources are available to the assistant.
- An MCP client is the AI assistant that sees those tools and calls them when it needs them to answer.
Write a server for your CRM once, and any compatible assistant can work with it, with no separate integration for each model.
Business examples
- A manager asks: “Which of client N’s orders are delayed and why?” The assistant checks Magento and the delivery service.
- An analyst asks to compare category sales for the quarter. The assistant runs a read-only query against the reporting database.
- Support gets a draft reply with the current order status and a returns link.
Security first
MCP gives AI access to real data, so design the server like a public API:
- least privilege: a dedicated service user, read-only by default;
- changing actions (refunds, price changes) only with explicit human confirmation;
- a log of every tool call;
- protection against prompt injection: data from emails, reviews and comments must never turn into commands;
- no secrets and no more personal data in responses than necessary.
We build MCP servers in PHP/Symfony and Node.js for CRMs, ERPs and Magento. Shall we discuss your use case?
