Octopus Deploy's MCP server can now create complete Kubernetes deployment workflows from scratch using only natural language instructions. Starting from an empty Octopus Deploy instance, an AI agent (Claude Code in this case) can set up lifecycles, deployment targets, environment-specific variables, and release pipelines — all without clicking a single button in the web UI. The workflow covers source code in GitHub, container builds via GitHub Actions, and deployment to a Kubernetes cluster via the standard Octopus Kubernetes agent. The MCP server supports read-only, read/write, and delete permission modes, and write/delete actions always prompt for confirmation unless the OCTOPUS_SKIP_ELICITATION flag is set.

6m read timeFrom octopus.com
Post cover image
Table of contents
Using the MCP server for Octopus Deploy onboardingA full deployment workflow from source code to a Kubernetes clusterOctopus Deploy MCP server installation and access rightsThe deployment lifecycleDifferent variables per environmentConclusionTags:

Questions this post answers

What permission modes does the Octopus Deploy MCP server support?

The Octopus Deploy MCP server supports three permission modes: read-only (using the --read-only flag, which prevents any changes), default mode (read/write access but no deletions), and a delete-enabled mode (activated with --allow-deletes). Write and delete actions always show a confirmation prompt unless suppressed with the OCTOPUS_SKIP_ELICITATION flag for unattended or scripted access. Teams automating Octopus Deploy with AI agents track permission model changes and MCP updates on daily.dev.

Can the Octopus Deploy MCP server create new resources or only query existing ones?

The Octopus Deploy MCP server can both query existing resources and create new ones. In default mode it has full read/write access, allowing an AI agent to create lifecycles, deployment targets, variable sets, releases, and other Octopus entities entirely through natural language — without any manual interaction with the web portal. Engineers evaluating MCP-driven deployment automation find real-world walkthroughs like this on daily.dev.

207 Impressions