We have just open-sourced Infobip OpenAPI MCP – a framework that helps transforming HTTP APIs based on the OpenAPI Specification into MCP servers. This is the same tooling that powers Infobip’s own MCP servers, which now is available for anyone building in this space.
Why We’re Releasing This
The Model Context Protocol ecosystem is growing rapidly, and we believe open tooling accelerates that growth for everyone. Here’s our thinking:
Contributing to shared infrastructure. MCP is still young, and the community benefits when companies share the building blocks they’ve developed. We’ve solved some tricky problems like adapting varied HTTP input patterns such as query parameters and headers into the JSON arguments that MCP tools expect, or transforming discriminators for polymorphic models into valid JSON schema. Rather than have others reinvent these solutions, we’d prefer to put them out there.
Growing together. The more MCP servers exist, the more useful the protocol becomes for developers and AI agents alike. By lowering the barrier to creating MCP servers from existing APIs, we hope to see more teams experiment, build, and contribute back.
Improving through collaboration. Open-sourcing invites feedback, bug reports, and contributions from developers who encounter edge cases we haven’t seen. That makes the tool better for everyone, including us.
What It Does
The Infobip OpenAPI MCP framework bridges OpenAPI specifications to MCP servers, built on Spring AI and designed to run as a Spring Boot application. Key capabilities include:
- Mapping HTTP API operations (paths and methods) to MCP tools
- Automatically wrapping and unwrapping inputs to match MCP’s expected format
- Transforming advanced OpenAPI schema features (like discriminators) into MCP-compatible JSON schema
A Note on Security
Because this framework exposes APIs to LLM agents, trust matters. Make sure you’re confident in both the API responses and the OpenAPI specifications you use—prompt injection risks apply here just as they do elsewhere. For third-party specs, consider storing reviewed versions as static files in your application.
If you’re working with OpenAPI-documented APIs and want to make them available to AI agents via MCP, we hope this framework saves you some time. And if you build something interesting with it, we’d love to hear about it (✉️ [email protected])!
