AI agent here means an LLM client (Claude Code, Cursor, etc.) driving the control plane. This is not the install runner in customer clouds.
Prerequisites
nuon agents context prints your auth, selected org/app/install, and MCP HTTP URL.
Connect
Recommended: the CLI stdio proxy forwards to the control-plane MCP server and injects your token and org ID from~/.nuon. Read-only by default.
--allow-writes on that command (or in args) to expose tools whose descriptions start with WRITE OPERATION:. Those tools also need a token with create permission.
Direct HTTP: point an MCP HTTP client at the URL from nuon agents context. The server is stateless Streamable HTTP (POST-only, no durable session), so auth and org go on every request as headers:
nuon orgs api-tokens create --name <name>. Because both paths set the org, you do not need select_org — only call it on HTTP without the header, and note it is a write tool.
Tools
Writes are hidden from the stdio proxy unless--allow-writes is set. HTTP MCP lists them whenever the token can create.
Common queries
Copy these into your LLM client after MCP is connected. Call tools in the order given. Write sequences need--allow-writes on the stdio proxy.
What’s in this org?
What’s in this org?
list_apps → list_installs (optional: filter installs with app_id)Status of install X?
Status of install X?
get_install → list_install_components → list_workflowsAny pending approvals?
Any pending approvals?
get_pending_approvals → get_workflow (use approval_id from the response)Approve or reject a pending plan
Approve or reject a pending plan
get_workflow → approve_step or reject_step (write)Why did this deploy fail?
Why did this deploy fail?
list_deploys → get_deploy → get_deploy_logs / get_workflow_step_logsWhy did this build fail?
Why did this build fail?
list_builds → get_build → get_build_logsWatch a workflow until it finishes
Watch a workflow until it finishes
watch_workflow (loop with last_known_status until the status is terminal)What actions can I run on this install?
What actions can I run on this install?
list_install_actions → get_actionRun an action on an install
Run an action on an install
get_action → run_action (write)Show runbooks for an app
Show runbooks for an app
list_runbooks → get_runbookOverview of an app branch
Overview of an app branch
list_app_branches → get_app_branchReturns answers.last_run_succeeded, answers.change_summary (config/git), and answers.deployment_summary (per install-group progress). Requires the app-branches org feature.Preview this PR (or git ref) against an install
Preview this PR (or git ref) against an install
list_app_branch_preview_sources → preview_app_branch (write) → watch_workflow → get_app_branchSame as nuon apps branches preview. Pass pr_number or git_ref plus install (name or ID). Default mode is plan-only; ask before apply.HTTP MCP cannot read the local workspace. For local toml, nuon apps sync first then pass app_config_id.Deprovision or reprovision an install
Deprovision or reprovision an install
get_install → deprovision_install / reprovision_install / deploy_install_components (write) → watch_workflowDeprovision requires confirm=true (ask the user first). plan_only does not require confirm.Inspect or update install inputs
Inspect or update install inputs
get_install_inputs → update_install_inputs (write)Retry a failed workflow step
Retry a failed workflow step
get_workflow → retry_step (write)Cancel an in-progress workflow
Cancel an in-progress workflow
get_workflow → cancel_workflow (write)Related CLI flags
These apply tonuon commands, not MCP tools:
See CLI commands for the full reference.