[On-Demand Webinar]
From Chat to Deployment:
AI Agents for Argo CD and Kargo
AI agents for Argo CD and Kargo: watch Claude inspect, sync, and promote applications through Akuity MCP, using the caller's own permissions.
Jiacheng Xu, Staff Software Engineer and team lead for AI at Akuity, and Eddie Webbinaro, Head of Solutions Engineering, demo the Akuity Agentic Control Plane (Early Access for selected Akuity Cloud organizations). Eddie covers how the Akuity Platform deploys, promotes, and operates applications, and how agent access is governed. Jiacheng then works from a Claude session: fleet insights across Argo CD instances, a selective sync blocked and then allowed by MCP guardrails, and a Kargo promotion risk assessment delegated to Akuity's built-in Promotion Advisor.
Read the Agentic Control Plane announcement → https://akuity.io/blog/agentic-control-plane-mcp-server
Getting started with Kargo: how GitOps promotion moves releases from dev to staging to production with Argo CD.
Taylor Thomas, Staff Software Engineer and Kargo maintainer at Akuity, walks through the GitOps basics behind Kargo, the open-source promotion tool built by the creators of Argo CD. He shows where Kargo sits between CI and Argo CD, defines warehouses, freight, and stages, then demos a dev, staging, and prod pipeline with verification, a pull request approval gate, and a rollback to earlier freight. Community Q&A covers replacing GitHub Actions scripts, two-registry setups, per-service Helm chart promotion, soak time, preview environments, and promotion windows in Kargo 1.12
Key Takeaways:
-
One Akuity MCP endpoint reaches every Argo CD and Kargo instance in an organization. Turn on the platform endpoint under MCP access in the organization settings, and a connected agent can list applications, health, sync status, images, and freight across all instances. Instance endpoints exist for teams that only want to expose a single Argo CD or Kargo instance.
-
Agents act as the signed-in user, not a shared service account. The agent authenticates through the user's SSO and inherits that user's permissions, modeled as actor, verb, and resource. If a user can't sync a production app in the UI, the agent can't either.
-
MCP guardrails add a second limit on direct MCP tools. In the demo, Claude's first sync attempt failed because the guardrail was set to read only. After Jiacheng raised it to read/write, Claude synced the out-of-sync dev and staging guestbook apps. Changes to managed state made through MCP are attributed and logged, so the record shows both the user and that the change came through MCP.
-
External agents can delegate work to Akuity's built-in AI agents. The open source Akuity agent plugins ship two skills: onboarding (create Argo CD and Kargo instances, connect clusters, bootstrap a GitOps repo) and delegate. Claude used the delegate skill to hand a prod promotion risk assessment to Kargo's Promotion Advisor, which has freight, verification results, and Argo CD health context. It returned a medium-risk assessment, and the promotion ran after review. Delegated work follows the built-in agents' own tool policies, separate from MCP guardrail levels.
-
Fleet insights include images and CVEs. Agents can pull data from Akuity Insights through MCP, such as the most-used images across all clusters and the CVEs for a selected image, without querying each instance separately.
Helpful Links:
- Agentic Control Plane announcement → https://akuity.io/blog/agentic-control-plane-mcp-server
- Agentic Control Plane governance docs → https://docs.akuity.io/akuity-portal/agentic-control-plane/governance-controls
- Fleet Management for Kargo Enterprise → https://akuity.io/blog/announcing-fleet-management-for-kargo-enterprise-one-release-hundreds-of-targets
- Kargo documentation → https://docs.kargo.io
- Book a demo → https://akuity.io/get-in-touch
- Join the Akuity & Kargo community on Discord → https://discord.com/invite/Y6Fs6HMZVR

