Kong Inc. today made available a technology preview of a registry for Model Context Protocol (MCP) servers and associated artificial intelligence (AI) tools to the catalog it makes available within the Kong Konnect application connectivity platform
The Kong MCP Registry adds a directory to register, dynamically discover and govern MCP servers within the context of a connectivity platform that is used to discover and govern other types of application programming interfaces (APIs) via the Kong API Catalog and a previously launched Kong AI Gateway, says Kong CTO Marco Palladino.
Compatible with an interoperability framework being advanced by the AI Alliance consortium, the Kong MCP Registry provides a centralized system of record for approved internal and external tools used by AI agents. Those AI agents will be able to simply query a catalog to determine which authorized tool best suits the task they’ve been assigned to autonomously complete, notes Palladino.
That capability enables IT teams to track dependencies, ownership, blast radius, and inherited policies by linking MCP servers directly to the APIs that they are built on, says Palladino. The overall goal is to make MCP servers more accessible versus having to constantly manually configure for each use case, he adds.
As more AI agents are deployed, organizations are realizing they lack the connectivity framework needed to enable them to automate workflows, notes Palladino. In the absence of that framework organizations wind up with a series of AI agents that are automating a limited range of tasks in isolation from one another, he says. “There’s too much fragmentation,” says Palladino.
Longer term, organizations will also need that connectivity frameworks to address everything from observability of AI agents to understanding the business value they generate to enabling monetization of those interactions, he adds.
Ultimately, AI agents will replace traditional user interfaces and even entire websites that were originally designed to enable humans to navigate content that soon AI agents will simply interrogate each other to discover and surface, notes Palladino.
It’s not clear how long it might take to build out the underlying infrastructure that will be required to enable AI agents to interoperate at that level, but it’s already apparent AI agents will be able to autonomously perform a wide range of tasks. In fact, Kong is working toward enabling AI agents to automatically invoke its tools and frameworks, says Palladino.
Of course, the fundamental economics of an agentic AI ecosystem have yet to be defined. Nor is anyone quite sure what impact the ability of an AI agent to perform a task is likely to have on capitalism if and when the day comes that most tasks are assigned to AI agents. The one thing that is certain is there will be a need to understand how those tasks were performed in a way humans can easily digest.
In the meantime, the race to provide the connectivity frameworks that AI agents require to manage tasks at scale is now clearly on.
