Stacklok, founded by Kubernetes veterans Craig McLuckie and Joe Beda, pivoted from software supply chain security to Kubernetes-based agent infrastructure after raising $17.5 million in a 2023 Series A from Accel, Madrona and Bain Capital. The company is now focused on enabling enterprises to run and govern agentic workloads at scale.
Mecatl: an open-source harness designed for the cloud
The company’s most notable project is Mecatl, launched as an open-source GitHub project in June. The name is pronounced "MEH-kah-tl" and is an Aztec word meaning a cord or rope. Mecatl bills itself as a "cloud-native harness" intended to move code and context out of local desktops and into centrally managed cloud infrastructure so enterprises can operate and govern agent interactions.
Why bring agents to the cloud?
Many early coding agents started as terminal tools or desktop apps; for example, Claude Code began life as a CLI tool. Those desktop-first harnesses tend to couple the agent loop, local execution and session state in the same process, which complicates enterprise management, session reliability, tool isolation and context preservation. McLuckie and Beda argue that a cloud-native approach enables rethinking why agent loops are so tightly bound to tool-calling subsystems or why agent identity is often handled through human identity systems.
Architectural principles
A central tenet of Mecatl is separating the agent loop from the client, model provider, state store and execution environment. The control loop is modeled as an application designed to run well in the cloud, while sensitive operations — tool invocation, shell execution, and session memory — are isolated into separate components. Session management and memory are moved away from local JSONL files into manageable systems that enterprises can operate and govern.
This decoupling is intended to let the harness scale beyond single-user, local deployments to infrastructure that enterprises can centrally operate. Beda compares it to enterprise email: interactions with agents while working for an organization should belong to and be governed by that organization.
ToolHive, MCP and the LLM gateway
Stacklok also developed ToolHive, an open-source platform for running and governing MCP servers. ToolHive started as a desktop, Docker-based way to manage MCP servers, then expanded into a Kubernetes-based gateway and related services — including a registry and an operator helper to manage MCP servers and handle input/output authentication.
MCP was the first step; the next is an LLM gateway (branded AI Gateway on Stacklok’s site) to control access, budgets and provider routing. The AI Gateway has not yet been open-sourced but is on Stacklok’s roadmap. Unlike some solutions that automatically pick a model per task, Stacklok’s gateway focuses on access control, budgeting, reporting and provider routing. McLuckie says semantic routing and model choice are better handled by an intelligent system embedded in the harness, which has more contextual information.
Reducing dependence on frontier labs and hyperscalers
A key motivation at Stacklok is helping large organizations reduce reliance on frontier labs and hyperscalers (examples cited in the industry include Codex, Claude Code and GitHub Copilot). McLuckie draws a parallel to Kubernetes’ role in decoupling workloads from specific cloud providers: the goal is to run agent workloads in a way that is less dependent on a single provider’s specifics.
Commercial model: the enterprise spine
With ToolHive and Mecatl as open-source components, Stacklok’s commercial offering centers on providing consistent identity, authorization, policy and auditing across those pieces — what McLuckie calls the "enterprise spine" or control plane that ties the open-source projects together.
Stacklok cites a survey of 520 leaders responsible for their organization’s use of large language models and/or AI agents to underline the scale and governance challenges. Beda notes that while open-source projects are relatively easy to start for small teams, new problems emerge when scaling across multiple clusters and clouds — problems Stacklok aims to solve.
Why this matters now
The article frames the effort as part of a broader pattern: early cloud services like Amazon S3 and EC2 were initially adopted by startups and individuals before enterprise-focused solutions and orchestration (notably Kubernetes in 2014) enabled large-scale, managed deployments. A similar transition may be occurring with agentic technologies: enterprises want centrally managed, auditable, scalable harnesses rather than desktop-first or small-team tools.
Stacklok — led by two architects of the cloud-native era — is betting that moving agent harnesses fully into the cloud, with Kubernetes-native architecture and a commercial control plane, will address enterprise needs for scale, governance and reduced vendor lock-in.
Key points
- Stacklok was founded by Craig McLuckie and Joe Beda and raised $17.5 million in a 2023 Series A.
- Mecatl is an open-source, cloud-native harness launched in June that separates the agent loop from execution and state storage.
- ToolHive provides an MCP platform and a Kubernetes gateway; Stacklok is also building an LLM/AI Gateway for access control and cost management (not yet open source).
- The company’s commercial product is an "enterprise spine" control plane that adds identity, policy and auditing across open-source components.
By shifting harnesses to a cloud-native, Kubernetes-aligned model, Stacklok aims to make agentic development workflows manageable and governable at enterprise scale.
