In this guest blog post, Josip Antoliš, Senior Principal Engineer at Infobip, discusses how MCP (Model Context Protocol) servers available in Microsoft Marketplace empower organizations to connect AI agents across multiple communication channels, driving richer customer engagement and operational efficiency.
As organizations race to become AI-first enterprises, a critical question emerges: How do you break your AI agents out of the prison of a chat box in a sidebar of your website and let them meet your users in conversations they are having in WhatsApp, Viber, or RCS?
The answer lies in MCP servers, an emerging standard that bridges the gap between AI agents and the real world of data and services, including enterprise communication platforms. For independent software vendors (ISVs) and Microsoft partners looking to deliver AI-powered solutions through Microsoft Marketplace, understanding this architecture is essential.
The business case for AI-enabled communication
Customers expect seamless, intelligent interactions across every channel. Whether it's a Microsoft Copilot-powered assistant scheduling appointments via SMS, a Microsoft Azure AI agent sending order confirmations through WhatsApp, or an automated workflow dispatching real-time alerts, it all boils down to one thing: AI agents need reliable, secure pathways to reach end users.
The challenge? Most enterprise communication platforms expose APIs designed for developers, not AI agents. The schemas, authentication patterns, and endpoint structures that work well for traditional integrations can create friction when AI needs to interact with them autonomously.
MCP servers solve this by acting as an intelligent translation layer, exposing communication capabilities in a format AI agents can understand and use reliably.
The Rise of MCP Servers: Enabling seamless AI access to business systems
For software companies building on the Microsoft cloud, MCP servers open significant opportunities:
Faster time to value with Azure AI
By wrapping communication APIs in MCP-compliant interfaces, partners can integrate Azure OpenAI Service and Azure AI Agent Service directly into customer engagement workflows. This means AI agents built on Microsoft infrastructure can send messages, manage conversations, and trigger notifications without custom integration code.
Simplified enterprise adoption
Organizations already invested in Microsoft 365 and Dynamics 365 can extend their AI capabilities to include external enterprise service, like omnichannel communication, all within a familiar, trusted environment. Procurement teams evaluating AI apps and agents can confidently assess solutions that leverage existing Azure infrastructure.
MACC-eligible communication solutions
For customers with Microsoft Azure Consumption Commitment (MACC), services like transactable communication solutions in Microsoft Marketplace provide a streamlined path to adoption. No new vendor onboarding, no non-standard payment terms, just seamless procurement that counts toward existing cloud commitments.
Solutions that combine AI agent capabilities with enterprise communication built on Azure infrastructure and available as transactable marketplace offers address a growing market need.
As Michal Pisarek, CEO of Orchestry, shared in a recent Partner Spotlight interview, typical procurement through Marketplace has compressed from around 120 days to under 30.
Architecture essentials: Building MCP servers that scale
Infobip, a global leader in omnichannel communication, has launched MCP servers that expose enterprise-grade messaging capabilities to AI agents.
Infobip’s MCP servers, available through Microsoft Marketplace, are architected on top of a robust API layer, ensuring flexibility and scalability for enterprise messaging. This approach allows seamless integration with existing platforms, while providing AI agents with direct access to messaging capabilities, not only sending messages, but also accessing customer data and creating customized engagement workflows.
Infobip believes this approach offers valuable lessons for partners considering similar implementations. Here are a few technical details of our solution:
Start with well-documented APIs as your foundation
Initial implementations mapped individual API operations to one MCP tool. Subsequent benchmarking and context engineering best practices revealed large language models (LLMs) perform optimally when provided with a targeted set of tools. Accordingly, Infobip split APIs across multiple smaller MCP servers and curated features that made sense for AI agents. We narrowed almost a thousand API operations down to a dozen MCP servers, each with a few essential tools.
A critical enabler for all this was OpenAPI specification. We had preexisting API governance in practice, ensuring every API operation is carefully documented in OpenAPI format, shares the same authentication and authorization, and uses unified API request and response models for common use cases like pagination. This enabled us to programmatically cherry-pick operations and features from our API product catalogue and compose them into MCP servers automatically.
Curate the AI experience
Starting with this strong API base, we focused on agentic experience. APIs used to be designed for human developers, and sometimes LLMs need special affordances. For example, we included a dedicated tool for resolving current date and time with every MCP server that enables scheduling of messages. With this simple addition, we’ve enabled LLMs that are notoriously bad at temporal concepts to send messages at precise dates in the future.
We’ve also employed prompt engineering techniques on metadata that introduces our MCP servers to LLMs. Those include instructions for using each server as a whole and descriptions of individual tools inside it. This is one more example where OpenAPI helps with its official overlays sub-specification. Overlays are a standard way to tweak a specification, and we use them to modify descriptions for best LLM performance.
Enterprise-grade security from day one
As previously highlighted, our API governance ensures the Infobip API benefits from industry-leading security standards. By building MCP servers atop our existing APIs, we naturally inherit these robust protections. Moreover, by staying aligned with the latest MCP specification updates, we've been able to deliver a seamless user experience.
In practical terms, users can access our MCP servers via an OAuth flow, which is automatically detected and initiated by modern MCP clients. The resulting bearer token is tightly scoped, restricting the AI agent’s access to only the specific features of the MCP server for which it was issued, preventing misuse. For server-to-server integrations, we also provide API key authentication, applying the same rigorously defined scopes to API keys as well.
Empowering developers with open source MCP platform
A lot of these features are helpful and reuseable on a general API scale. We built Infobip MCP servers on top of our OpenAPI documented HTTP API, but the same principle can be applied to other APIs as well.
This was the motivation behind our decision to open source our framework. It is based on Spring AI and enables developers to build Java-based MCP servers on top of a well-documented API.
The framework is easy to use and flexible. You can get started with a simple configuration for basic needs or customize and enhance it further using Java code if required. In addition to features for customers, the framework benefits from the Java platform’s strengths: It's portable, easy to maintain, and performs well. It also includes advanced, production-grade features, such as built-in open telemetry tracing for monitoring and diagnostics.
You can find the source code, documentation, and example projects in our GitHub repository.
Getting started with MCP
Whether you're an ISV exploring AI agent capabilities, a Microsoft partner expanding your marketplace presence, or an enterprise architect evaluating communication solutions, MCP servers represent a strategic opportunity to connect AI innovation with real-world customer engagement.
For partners
Consider how MCP-compliant communication capabilities could enhance your existing Azure-based solutions. Explore ISV Success for technical consultations, publishing support, and leverage Marketplace Rewards to accelerate your go-to-market strategy.
For enterprise buyers
Look for AI-enabled communication solutions in Microsoft Marketplace that offer transactable purchasing, MACC eligibility, and native Azure integration.
For instance, by connecting Copilot to MCP-enabled communication channels, organizations can upgrade their features to include automated customer messaging, proactive notifications, and even personalized engagement across various communication channels. This means Copilot can evolve from simply assisting with internal workflows to becoming a full-fledged omnichannel AI assistant, driving richer customer interactions and unlocking new business value.
For developers
Dive deeper into MCP server architecture through resources like the Infobip MCP documentation and the open-source Infobip OpenAPI MCP framework. Understanding this emerging standard positions you to build the next generation of AI-powered enterprise applications.
Infobip's MCP servers, which enable AI agents to communicate across WhatsApp, SMS, Viber, RCS, and more, are available through Microsoft Marketplace. Learn more about how partners are leveraging MCP architecture to deliver AI-powered communication solutions.
