Kubernetes creators launch cloud-native agent harness Mecatl to decouple agent loops from local execution

Stacklok, co-founded by Kubernetes creators Craig McLuckie and Joe Beda, launched Mecatl, a cloud-native agent harness designed to shift agent loops from local desktops into Kubernetes environments. The project began as an open-source initiative on GitHub in June and emphasizes treating agent infrastructure as cattle rather than pets, aligning with Kubernetes’ scalability and reliability principles.

Mecatl’s architecture decouples the agent loop from tool execution, client interfaces, state stores, and execution environments. This separation aims to improve reliability by isolating sensitive operations like tool calling and session management from the agent loop, which Stacklok argues is better suited for cloud-native deployment. The company frames this as moving the locus of value—such as code and context—from local environments into managed cloud systems, particularly for enterprises where intellectual property and governance are critical.

Stacklok also open-sourced ToolHive, a platform for running and governing MCP servers. Initially focused on desktop deployments, ToolHive has expanded into a Kubernetes-based gateway and registry to manage MCP servers at enterprise scale. The company positions ToolHive as a way for organizations already using Kubernetes to integrate agentic workloads into their existing infrastructure.

To address enterprise needs in regulated industries, Stacklok is developing an AI Gateway (LLM gateway) for access control, cost management, and provider routing. While not yet open-sourced, the company states it is on the roadmap. The gateway focuses on governance and budgeting rather than model selection, which Stacklok argues is better handled within the harness where more context is available.

The company’s broader mission is to help enterprises reduce dependency on frontier labs and hyperscalers by providing infrastructure that can operate across cloud environments decoupled from specific providers. This mirrors the original vision of Kubernetes, which enabled running applications across heterogeneous cloud providers without lock-in.