At the end of May 2026, Cloudera and vCluster brought platform engineers, cloud-native builders and AI infrastructure teams together in Koramangala for the AI Infrastructure Meetup Bengaluru.
NitroStack was there.
And looking at the agenda, there was an obvious overlap. The room was discussing many of the same production problems that NitroStack has already been turning into an MCP-native development stack.
Not just how to build with AI.
How to actually run it.
The agenda looked a lot like the MCP production stack
The sessions moved quickly past basic AI demos.
There were conversations around AI factories, Kubernetes-scale infrastructure, MCP and tool-enabled AI engineering, evals and tracing, MCP identity propagation, enterprise serverless agents, and production observability.
That combination matters.
Because once MCP moves beyond localhost, these are exactly the problems developers run into.
How does an MCP server scale? How is identity carried through a tool call? What happens when traffic spikes? Where do the logs go? How do you inspect what the agent actually did? How do you deploy without rebuilding an entire DevOps stack around every MCP application?
This is where NitroStack is particularly well positioned.
NitroCloud is already built around these problems
Take serverless infrastructure.
NitroCloud is purpose-built for deploying MCP servers, with scale-to-zero compute, automatic scaling, Git-based deployments, automatic containerization and production HTTPS endpoints.
Instead of asking an MCP developer to configure Dockerfiles, deployment YAML, rollout infrastructure and capacity planning separately, NitroCloud collapses that layer into the deployment workflow.
Then there is observability.
NitroCloud provides real-time logs and infrastructure monitoring across CPU, memory, network and request activity. NitroStudio goes further on the development side, giving MCP builders visibility into tools, resources, prompts, execution flows, latency and how AI actually interacts with the server.
So when observability became part of the conversation in Bengaluru, it wasn't an abstract future requirement for NitroStack.
It was already part of the product.
MCP identity is becoming infrastructure, not an add-on
One of the more relevant sessions at the meetup went directly into identity propagation in MCP, multi-hop chains and trust.
This is another area where production MCP gets serious very quickly.
NitroStack's stack already treats authentication and security as infrastructure concerns, with OAuth-oriented MCP development, encrypted secrets, JWT-based controls, RBAC, TLS and managed HTTPS available across the production layer.
The direction is pretty clear: MCP servers cannot become serious enterprise infrastructure if identity, permissions and security remain things developers bolt on at the end.
NitroStack is building for that reality from the beginning.
The broader AI infrastructure conversation is catching up to MCP
What made this event interesting for us wasn't discovering a new infrastructure thesis.
It was seeing just how closely the broader AI infrastructure conversation now maps to the problems NitroStack is already solving.
Serverless MCP deployment. MCP observability. Identity and security. Testing. Scaling. Production reliability.
Taken individually, these are infrastructure problems.
NitroStack's advantage is bringing them together around one MCP development workflow: build and inspect in NitroStudio, deploy and operate through NitroCloud, then ship the MCP server into real AI applications.
That puts NitroStack in a strong position as MCP development moves from experimental servers toward production infrastructure.
And being in Bengaluru with platform engineers, cloud-native teams and AI builders was another chance to put NitroStack directly inside that conversation.
Not watching it from the sidelines.
Building the stack for it.