Video
As you install MCP servers, the risk is that you start crowding your context window with all the tool metadata tied to those servers. While a few tools may be needed for a task, most of the metadata simply wastes tokens and risks degrading model performance. Craig McLuckie of Stacklok launched a simple solution to …
Nick Aldridge of AWS is one of the core maintainers of the Model Context Protocol. He joined MCP Academy to talk about the architecture of a multi-tenant MCP gateway, and then showed the audience exactly how to bring it to life with AgentCore Gateway.
There has been a lot of recent buzz about browser agents like ChatGPT ‘Atlas’ and Perplexity ‘Comet’ but there are very real security concerns with how they work today. Alex Nahas who created MCP-B highlight some of those concerns and then showed how WebMCP could be a mitigating force.
In the first half of 2025, Baxter Black from Madrona co-authored a post that broke down patterns and popularity in the MCP ecosystem. Just in time for MCP Academy, Baxter revisited the research to highlight which MCP servers are being used most widely and hinted at the likely use cases driving enterprise interest in MCP.
With increasing use of AI coding assistants there’s growing pressure on developers to maintain control and visibility of their code. Brock Lumbard from Statsig offered some guidance and then jumped into a demo that resonated with the developers in the audience.
Early at MCP Academy, Caitie McCaffrey of Microsoft took the stage to demo simple, effective ways to build your next MCP server. Whether you’re just starting your MCP journey, or you’ve already built a number of MCP servers, Caitie’s talk has you covered.
Most enterprises maintain API catalogs with documentation, specs, reference guides and more that is designed for human consumption. What happens when developers need their AI copilots to discover, understand and interact with APIs? Ross Kukulinski from Kong showed us how to use MCP servers to enable AI agents to conversationally explore available APIs.
The first session at MCP Academy was an important opportunity to get the entire audience aligned on a definition of the protocol and its potential. Eleftheria Stein-Kousathana from Stacklok set the tone with a clear outline of the Model Context Protocol.
There’s a joke that the ‘S’ in MCP stands for security. Working out how to build the right security controls is a critical obstacle for many production deployments. Jai Behl and Satish Sheth from Databricks share ideas and examples that make it real.
A lot of enterprises are experimenting with MCP, while a brave through are putting it into production. Harish Gaggar shares how Credit Karma constructed the necessary connectors and security guardrails to apply MCP in a regulated industry.
The MCP spec calls for a 1-to-1 relationship between client and server; however, there are situations where you need multiple tools from multiple MCP servers to complete a task. Dave Dixon from Credit Karma is pushing for a solution.
One of the hot button topics when it comes to MCP is observability. Sure, MCP can introduce new observability challenges, but it might also be able to resolve some long-standing observability issues. One of the more popular MCP servers is from Grafana, and Tiffany Jernigan showed the audience how to put it to work.
