Infobip's MCP Servers Enable Smarter AI Communication (shiftmag.dev)

At Infobip Shift in Zadar, Principal Engineer Filip Srnec unveiled Infobip’s MCP servers—an implementation of the Model Context Protocol (MCP) designed to give AI agents “communication superpowers.” The servers act as a universal communication layer that standardizes how agents access tools and channels (SMS, WhatsApp, Viber, RCS), call airline or messaging APIs, trigger 2FA, and pull customer insights from Infobip’s People CDP. For developers and the AI/ML community this matters because MCP promotes interoperability (switching AI providers without rearchitecting integrations), delivers richer contextual inputs to models, and abstracts channel and compliance complexity so agents can act reliably across real-world systems. Technically, Infobip maps MCP tools 1:1 to OpenAPI operations so MCP servers can dynamically discover and call endpoints based on OpenAPI specs—enabling scalable, auto-generated tooling. Filip highlighted practical challenges: mismatches between OpenAPI’s JSON Schema and broader JSON Schema vocabularies, the need to flatten OpenAPI discriminators into oneOf for simpler input matching, and reference-resolution complexities that require thoughtful API design to achieve the desired “Agentic Experience.” Security is handled by MCP servers acting as a secure proxy with middleware to verify credentials and support OAuth via metadata discovery. The result is a reference implementation and architecture that make agentic systems more practical, secure, and interoperable—demonstrated already in integrations like Autorun.ai.