Kong Releases Volcano, an MCP-Native SDK for Building AI Agents
NEW YORK — API and AI company Kong released a new TypeScript-based, open source software development kit (SDK) for creating AI agents at Tuesday’s API Summit 2025.
The SDK, called Volcano, is the first to take a Model Context Protocol (MCP)-native approach to creating AI agents, according to Mark Palladino, Kong CTO and co-founder.
During his keynote presentation on Tuesday, he demoed an AI agent powered by Volcano by having audiences scan a QR code. That pulled up an interface, which asked for the participant’s name and location. Based on the information, it auto-generated an order to a coffee shop located near the conference. In real time, it placed the coffee orders — over 900 orders.
It did that in 15 lines of code, Palladino told the audience.
Why Kong Developed an MCP-First Framework
Volcano stands out over older solutions in that it is MCP-first, he added. It’s otherwise similar to solutions such as LangChain, Haystack and Flowise.
“There are many frameworks out there and many frameworks were actually invented, relatively speaking, quite a long time ago in AI,” Palladino told The New Stack. “Most of them come up with a certain level of complexity and number of abstractions that make it quite hard to actually build agents in a beautiful, concise way.”
Kong developers learned this the hard way after trying a variety of these libraries. Most solutions were built before MCP was created by Anthropic last year, or at least before it became a widely-embraced standard, he added.
The development team at Kong thought there must be a better way to build AI agents, so they created the framework to use internally.
The Role of the Model Context Protocol (MCP)
MCP has emerged as the connectivity layer in the agentic AI space, he continued. It connects AI to data sources, systems and services, allowing AI agents to leverage these resources.
“Now that MCP has emerged as the underlying, foundational connectivity layer in the agentic AI space, obviously, we built Volcano with MCP-native capabilities in mind,” he said.
It supports it natively, instead of adding MCP in a way that seems like it’s added onto something that’s already existed for years, he added.
“Given that we at Kong are also building lots of agents; and, by the way, we tried all of these frameworks, we thought that there was a better way to do it,” Palladino said. “That’s why we built Volcano for ourselves, first and foremost.”
Volcano “has been born with MCP in mind.”
– Mark Palladino, Kong CTO
Developers don’t have to use Volcano to build their agents — any solutions will work with Kong’s AI Gateway, he said.
But Kong believed Volcano is a better tool, so the company — which has a history built on open source — decided to give back to the community by open sourcing Volcano.
“Open source developers have yet another option they can choose from to go and build MCP native agents,” Palladino said. “And arguably, for MCP native agents, Volcano is probably the best choice because it has been born with MCP in mind.”
Key Features and Capabilities of Volcano
Volcano supports thousands of models out of the box from a variety of vendors, including Anthropic, OpenAI, Azure, AWS and GCP.
But what makes Volcano exciting is that it can create flows, Palladino said. Each part of the flow can potentially use a different LLM.
“For example, making a request that requires advanced reasoning, I may want to use a model from my vendor that is very good for advanced reasoning; but then for the next flow, the next step into that workflow, I could be using another LLM,” he said. “We can automatically detect the right tools within that MCP server that we should be using in order to be able to make the best request to those MCP so it’s fully MCP native.”
That can help developers, who sometimes only find the right model for their project through trial and error. Volcano can also create subagents that can be consumed by other agents natively via the code, he said. The plan is to continue adding support for more models, creating an ecosystem of integrations to make it easier to build with Volcano, he said.
While Kong plans to support the open source project, it is looking for maintainers. This week’s Volcano announcement was a soft launch. Next week Kong will be adding tutorials and other supporting features for Volcano.
New MCP Capabilities in Kong’s AI Gateway
Kong CEO and co-founder Augusto Marietti also unveiled new MCP gateway capabilities to Kong’s AI Gateway, which launched last year. Kong can automatically generate MCP servers from Kong-managed APIs.
He also revealed that Konnect Developer Portal is now MCP-enabled, which gives developers a way to build, run, discover and govern MCP-based tools and agentic AI-powered applications.
MCP capabilities inside the AI Gateway include the ability to automatically generate secure MCP servers from Kong-managed APIs, essentially MCP-enabling APIs so that they can be used with AI agents.
Authentication and authorization can be a challenge with AI agents, so Kong has introduced a centralized MCP authentication via its dedicated MCP OAuth 2.1 authentication plugin. It can validate tokens and apply consistent security policies across every server automatically.
Finally, Marietti also introduced an AI Integration Composer for discovering, composing and mapping intelligent API endpoints as tools for MCP servers, which he noted marked the company’s entrance into AI-based integration.
