Person
1 episodes · 24 quoted claims
Craig McLuckie says he and a colleague named Joe built Google Compute Engine, Google's virtual machine infrastructure service.
“I was, uh, happy to meet my friend Joe and we built out, uh, Google Compute Engine, which is, you know, virtual machine infrastructure technology.”
McLuckie says he and his colleague started the Kubernetes project while at Google and later built other technologies together.
“We started the Kubernetes project, did a few other things, you know, inside there, we, you know, subsequently, we've built a lot of other, uh, technologies together.”
Docker is a technology that packages an application and its dependencies into a single portable container.
“Docker is a technology. If if if the audience isn't familiar with it, that enables people to package up an application and all of its dependencies in a single container”
Anthropic created the Model Context Protocol, which describes outside tools in simple natural language backed by a JSON schema so language models can use them.
“what Anthropic did was start to describe the outside world in relatively simplistic natural language terms. Uh, with a JSON schema backing it.”
McLuckie says an MCP platform has four pieces: a runtime, a registry, a gateway and a control plane.
“I think of as being four pieces that go into that. The 1st piece that goes into that is I need a runtime, I need somewhere where I can run this thing”
McLuckie says developers commonly run MCP servers by pulling a package from the internet with npx and running it locally, which risks running exploited code.
“In the case of, you know, most development technologies, people go npx run. They basically just pull a package down from the internet and it runs locally.”
McLuckie says his CTO Joe wrote the SPIFFE paper, about a zero-trust identity framework, around ten years earlier.
“He wrote the Spiffy paper, I don't know if you've heard of Spiffy. It's it's an identity system that's kind of, uh, a sort of, the sort of zero trust”
McLuckie says the MCP specification was grounded in OAuth 2 workflows, where a user's OIDC token identifies them to an MCP server.
“MCP as a specification was really grounded in, like O, O, O, two workflows.”
McLuckie describes agent identity as a three-legged stool: a service account identity for the agent, claims from the agent owner's role, and on-behalf-of claims inherited from the user.
“There's what I think of as a service account identity that's effectively identifying the agentic endpoint.”
McLuckie gives the example of configuring the AWS MCP server in read-only mode by exchanging an Okta token for a scoped AWS token, to prevent an agent from deleting an RDS instance.
“I want to use the AWS MCP server and I want to use in read only mode, right?”
McLuckie says agent-only authorization policies can be written as policy-as-code using technologies such as Cedar or Rego and applied through a common proxy.
“you can start to describe those in a in a technology like Cedar or Rego or what have you.”
McLuckie says pulling in three or four MCP servers can create around 150 tools and use 20-30k tokens on each interaction.
“you may have 150 tools. You may be burning 20 30k tokens every every interaction.”
McLuckie says replacing many tool listings with find-tool and invoke-tool endpoints reduces input token consumption by 80 to 90%.
“It reduces input token consumption by eighty to 90% when you have these these things, and so that's a very big deal.”
McLuckie says a single guided endpoint can raise tool-calling reliability by smaller models back to roughly 95-97%.
“You can get back up to the sort of ninety five, ninety seven threshold that actually makes the system useful.”
ToolHive is an open source project licensed under Apache 2.
“ToolHive is an open source project, it's Apache 2 licensed.”
ToolHive packages MCP servers in Linux application containers delivered as OCI images that enterprises can scan, harden and validate.
“let's just put our MCP servers in a Linux application container. You know what that means? Well, for an enterprise, that means that, uh, it's an OCI image.”
Containers let ToolHive restrict which parts of the file system and which network endpoints an MCP server can access.
“So I can describe which portions of the file system it can access, I can describe what network endpoints it can access.”
McLuckie says ToolHive's use on Kubernetes is growing about fifty percent month over month.
“we're seeing, you know, fifty month over month growth in the Kubernetes use of this technology”
McLuckie says ToolHive is receiving millions of tool invocations from Kubernetes deployments.
“we're getting millions and millions and millions of of of of tool invocations from the the Kubernetes side of the house.”
McLuckie predicts AI systems will generate Kubernetes YAML manifests from descriptions and monitor deployments, stepping in when ordinary reconciliation fails.
“you'll be able to, you know, describe what you want to have happen, you know. It'll basically generate the YAML manifest, hand it off to Kubernetes”
McLuckie says Stacklok instruments its developers' agent use with hooks to track which agents they use and how.
“all of the developers have, you know, hooks. We know exactly which agents they're using, how they're using”
McLuckie says Stacklok developers run between five and 15 concurrent agents, each configured with a different role and controlled tool access.
“they'll have 15 different agents, each with a slightly different configuration recipe role, uh, performing a set of tasks”
McLuckie says Stacklok's engineering team throughput rose 60 in one week as the team improved at running agents concurrently.
“This last week, our person like our, our engineering team's throughput went up 60 in a week.”
McLuckie says Stacklok can now close community issues faster than new ones arrive.
“we're finally over the threshold where we're actually able to burn down issues faster than they're coming in.”