{"id":3667,"date":"2025-10-03T08:46:45","date_gmt":"2025-10-03T08:46:45","guid":{"rendered":"https:\/\/www.infobip.com\/developers\/?p=3667"},"modified":"2025-12-17T13:15:47","modified_gmt":"2025-12-17T13:15:47","slug":"enabling-agent-communication-how-to-build-an-mcp-server-on-top-of-an-api","status":"publish","type":"post","link":"https:\/\/www.infobip.com\/developers\/blog\/enabling-agent-communication-how-to-build-an-mcp-server-on-top-of-an-api","title":{"rendered":"Enabling Agent Communication. How to Build an MCP Server on Top of an API"},"content":{"rendered":"\n

MCP (Model Context Protocol) servers are quickly becoming a crucial layer between software systems and AI agents.  In the context of systems that already expose a public API, MCP servers can be positioned as proxies, translating standardized API specifications into structured data that AI can easily consume. Done right, MCP servers allow AI agents to interact with external platforms, access data, trigger workflows, and perform complex operations\u2014all while staying decoupled from the underlying system. <\/p>\n\n\n\n

But here\u2019s the catch: building MCP servers isn\u2019t just about wrapping your API in a shiny new layer<\/strong>. If you do it naively, you\u2019ll expose too much, risk security, and run into schema nightmares. <\/p>\n\n\n\n

In his talk at the Shift Zadar 2025<\/a> conference, Filip Srnec, a Principal Engineer at Infobip, shares how Infobip approached the topic of creating their MCP servers, focusing on OpenAPI, schema design, and tricky features that can make or break your implementation. <\/p>\n\n\n\n

\"Filip
Filip Srnec giving his talk, How Can Agents Communicate. Unlocking AI Interactions with Infobip MCP Servers.<\/figcaption><\/figure>\n\n\n\n

The Core Idea: MCP as a Decoupled Layer<\/strong> <\/h2>\n\n\n\n

At its heart, an MCP server is another interface<\/strong> to your platform. AI agents can connect to the MCP layer. And in turn, the MCP layer can communicate with your core API stack. This means MCP can be fully decoupled<\/strong>\u2014not just a wrapper glued onto the existing endpoints. <\/p>\n\n\n\n

To scale the MCP servers, the natural question arises: can we deploy them dynamically from APIs? The answer is yes\u2014 by leveraging the <\/strong>OpenAPI specification<\/strong><\/a>.<\/strong> <\/p>\n\n\n\n

OpenAPI: Your Blueprint for MCP Servers<\/strong> <\/h2>\n\n\n\n

OpenAPI gives you a formal, machine- and human-readable way to describe HTTP APIs. <\/p>\n\n\n\n

It\u2019s the perfect foundation for MCP because it provides a clear and standardized way to translate API endpoints into tools that AI agents can understand and use. <\/p>\n\n\n\n

\"MCP
Official MCP tool data type specification<\/a><\/figcaption><\/figure>\n\n\n\n

The goal is one-to-one correspondence: every MCP operation should directly map to an API operation. <\/p>\n\n\n\n

\"MCP
An example of how to match API operations to the MCP tool data type specification.<\/figcaption><\/figure>\n\n\n\n

But this doesn\u2019t mean you should just expose your entire API spec<\/strong> to AI. That could result in suboptimal agentic experience. <\/p>\n\n\n\n

Filip stresses this point several times in his talk: build your MCP to match the AX you are after<\/strong>, rather than throwing everything you\u2019ve got at it.<\/p>\n\n\n\n

\n

Curate what\u2019s exposed, rewrite descriptions with context, and even restructure certain endpoints. <\/p>\n<\/blockquote>\n\n\n\n

The Discriminator Problem<\/strong><\/h2>\n\n\n\n

According to Filip, one of the most powerful\u2014but also challenging\u2014parts of OpenAPI is the discriminator<\/strong>. <\/p>\n\n\n\n

MCP tools use standard JSON Schema, while OpenAPI<\/a>\u2019s version is a superset that includes features such as discriminators. We therefore need to translate discriminators into the standard JSON Schema format as effectively as possible. <\/p>\n\n\n\n

Let\u2019s say your API, just like Infobip\u2019s, supports sending messages using Viber or RCS. A single endpoint might handle multiple message types: text, images, video, files. The discriminator lets you define a property  <\/strong>(e.g., “type”: “image”) that determines which input schema to use.<\/p>\n\n\n\n

\"Example
An example of how to use a discriminator with a defined property.<\/figcaption><\/figure>\n\n\n\n

In theory, this is elegant. In practice, it introduces several challenges: <\/p>\n\n\n\n