MCP explained: what the Model Context Protocol means for integration
MCP has become the standard way to connect AI agents to tools and data. What it is, how it relates to APIs and iPaaS, and what integration teams should do about it.
If you work with integration, you have probably heard “MCP” more often this year than in all previous years combined. The Model Context Protocol has gone from an Anthropic side project to something every major iPaaS vendor supports. Here is a short explanation of what it is and why it matters for integration teams.
What MCP is
MCP is an open standard for connecting AI applications to external tools and data. It was introduced by Anthropic in November 2024, and in December 2025 it was donated to the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. OpenAI, Google and Microsoft all support it.
MCP is often called “USB-C for AI”. Before MCP, every AI application needed its own custom connection to every system, the classic N×M problem integration people know well. With MCP, a system is exposed once as an MCP server with a set of tools, and any MCP client (Claude, ChatGPT, Copilot, a home-grown agent) can find and use those tools.
MCP and APIs: not either-or
An MCP server is usually a thin layer on top of existing APIs. The difference is who the consumer is. An API is designed for a developer who reads documentation and writes code. An MCP tool is designed for a language model that reads a description and decides for itself when to use it.
That has consequences. The tool description becomes part of the interface. Tools should be designed around tasks (“create a customer order”) rather than raw endpoints. And because the model decides what to call, access control and logging become even more important.
Why iPaaS vendors care
An integration platform already has what MCP needs: connectors to hundreds of systems, managed credentials, error handling and monitoring. So the vendors are turning their assets into MCP tools:
- Azure Logic Apps can expose existing workflows as MCP tools (generally available since Build 2026).
- Boomi has an MCP registry in its API management and MCP proxy connectors in Boomi Connect.
- Workato has an MCP gateway and an internal marketplace for reusable MCP servers.
- MuleSoft supports MCP in Agent Fabric and governs MCP servers through Omni Gateway.
Gartner also highlights MCP in its 2026 Magic Quadrant for iPaaS, as a way to connect AI to enterprise systems without locking yourself to one vendor.
The protocol is maturing
The latest specification, dated 28 July 2026, makes MCP stateless. The handshake and session IDs are gone, and each request describes itself. That makes it much easier to run MCP servers behind load balancers and gateways, the way we already run APIs. Routing headers, caching of tool lists and stricter authorisation rules point in the same direction: MCP is becoming ready for enterprise use.
What integration teams should do
- Start with what you have. Pick a few well-used integrations or APIs and expose them as MCP tools with clear, task-oriented descriptions.
- Govern MCP servers like APIs. Catalogue, version and put them behind your API gateway.
- Give agents their own identities. Do not let agents share service accounts, and give each the least access it needs.
- Log tool calls. You will need to be able to explain what an agent did, and why.
MCP will not replace integration platforms. But it changes who uses them: from now on, some of the consumers of your integrations will be AI agents.
Sources
This post was written with AI assistance and reviewed by the editor before publishing.