Skip to main content
The Astro Private Cloud (APC) Model Context Protocol (MCP) server exposes 37 tools, grouped into seven tool groups. A tool group is a Helm chart toggle: disabling a group removes its tools from the catalog entirely, not just from what a given caller can see. See Set up and configure the MCP server to enable or disable a group and for which groups are on by default. The always-on core group handles health and liveness checks and exposes no tools to callers, so it isn’t listed here.
Astro Private Cloud 2.2This feature was introduced in Astro Private Cloud 2.2. To access this feature, upgrade your Astro Private Cloud installation to 2.2 or later.
Each tool declares an effect (Read, Mutate, or Destroy) that a client can use to decide whether a call needs confirmation. Effect and group are independent: a few tools in the platform_write group (for example rollback_deployment) have a Destroy effect even though the group isn’t platform_destroy. Check a tool’s own effect rather than assuming it from its group. The MCP permission listed for each tool is the server’s own label, not an APC API permission, and doesn’t appear in APC roles. The APC API authorizes every call against the caller’s own role, so a call succeeds only when that role allows the underlying action. The server doesn’t prompt the user before a call. Destructive tools carry a destructiveHint annotation that a client can use to ask the user first, but the hint is advisory, and the agent decides whether to honor it. The server also requires confirm: true on the seven tools that take a confirm argument and rejects any call without it with a denied error, whatever the client does.

​
Platform read (platform_read, on by default)

Read-only platform operations. They report state and don’t restart Deployments.
No arguments.
No arguments.
Returns the Deployment’s base info, active URLs, config summary, and environment variables.
Use this to choose a revision for rollback_deployment.

​
Platform write (platform_write, on by default)

Deployment and Workspace writes. update_deployment, create_deployment, deploy_image, and rollback_deployment restart the Deployment’s Airflow components. Each call drives a Helm upgrade. Workspace membership and team tools don’t touch a Deployment.
Every argument besides deployment is optional. Always pass isSecret for an environment variable. Leaving it out stores the value as a plain variable, even if it was a secret before. A component object such as workers or scheduler replaces that component’s whole configuration, so pass every field, not just the one you’re changing. Each call triggers a Helm restart of the Deployment.
This uses the same Helm-apply path as update_deployment, run for the first time for this Deployment, so there’s nothing running yet to restart.
Re-renders the Deployment to a prior Helm release. This restarts the Deployment, and you can reverse it only by rolling forward again.

​
Platform disruptive (platform_disruptive, off by default)

Deployment contract changes: executor, runtime version, or Dag (directed acyclic graph) delivery mode. Kept separate from the everyday write group so enabling routine config updates never also enables contract changes.
A contract change, not a routine config update. Each call triggers a Helm restart, as update_deployment does.

​
Platform destroy (platform_destroy, off by default)

Deletes a Deployment or a Workspace outright. Neither tool restarts anything, and both tear down a Helm release rather than re-rendering one.

​
Airflow read (airflow_read, off by default)

Airflow diagnosis: task logs, import errors, installed providers, health, and Dag run detail. get_task_logs, get_import_errors, get_installed_providers, get_airflow_health, and get_airflow_component_health work on both Helm-based and operator-based Deployments. get_dag_stats, get_dag_run, and get_dag_run_diagnosis use the Laminar Airflow-state service and return results only for operator-based Deployments. check_upgrade_safety reaches neither the APC API nor a live Deployment. It screens a version against a vendored knowledge base.
Reports Dag-authoring health: Airflow version, import errors, Dag warnings, and Dag run stats by state, with an overall healthy, warning, or unhealthy status. Dag run stats are available on Airflow 3 only.
Reports process liveness for the metadata database, scheduler, triggerer, and, on Airflow 3, the Dag processor, each with a healthy or unhealthy status and its last heartbeat. Distinct from get_airflow_health, which reports Dag-authoring health.
Filters a run’s task instances to the failed and upstream-failed ones.

​
Airflow destroy (airflow_destroy, off by default)

Destructive Dag run and task-instance operations, in their own group so you can enable them independently of platform_destroy. All three call the Deployment’s Airflow REST API through the APC API, so they work on both Helm-based and operator-based Deployments. None restarts a Deployment.

​
Skills (skills_load, off by default)

Skill-delivery tools are read-only and carry no permission requirement of their own. They return vendored content, not platform state. astronomer.mcpServer.skills.enabled toggles them, and you don’t list them in enabledGroups. See Astro Private Cloud MCP server skills for how these combine with the per-skill prompts.
No arguments. Returns every loadable skill, each with its name, description, and origin.