Skip to main content

Palette MCP Server Operations

The Palette MCP server exposes read and write tools grouped by the Palette resource they act on. Write tools are gated behind the --allow-write startup flag and remain disabled by default. Delete tools additionally require you to type back the resource name to confirm.

AreaRead toolsWrite tools
Clustersread_clusters, read_cluster_status, read_cluster_observability, read_attached_profiles_to_cluster, read_cluster_kubeconfig, fetch_pre_cluster_create_schemacreate_cluster, update_cluster, delete_cluster
Cluster profilesread_cluster_profilescreate_cluster_profile, delete_cluster_profile
Cluster tagsread_cluster_tagsupdate_cluster_tags
Projectsread_projectscreate_project, update_project, delete_project
Teamsread_teamscreate_team, update_team, delete_team
Usersread_userscreate_user, update_user, delete_user
Packsread_packs—
Registriesread_registries—
Cloud accountsread_cloud_accounts—
Edge hostsread_edge_hosts—
Eventsread_events—
info

Cluster tag tools operate on clusters only. Tag management for cluster profiles, cluster templates, edge hosts, and cluster template policies is not available through the MCP server.

This guide provides examples of how to use the Palette MCP server using Claude Code and the Claude Sonnet 4.6 model. You can use an MCP client and model that suits your needs.

Prerequisites​

  • Ensure the following software is installed locally on your workstation:

    • The Palette MCP server configured and set up. Refer to the applicable setup guide for Claude, Cursor, or Antigravity.

    • A container engine, such as Docker or Podman, if you use the container image.

  • A Palette account.

  • A Palette API key. Check out the Create API Key guide for more information.

List Clusters and Cluster Profiles​

The read_cluster_profiles tool lists cluster profiles. The read_clusters tool lists clusters.

  1. Open your MCP client.

  2. Send a query asking for information about your Palette cluster profiles. For example, you can ask "Which Palette cluster profiles have the env:prod tag?" to learn which cluster profiles have been created with a particular cluster profile tag.

    Example Output
    ⏺ palette - read_cluster_profiles (MCP)(filters: { tags: ["env:prod"] })

    ⏺ Profiles explicitly tagged with env:prod:

    ┌──────────────────────┬─────────┬──────────────────────────────────────────────────────────────────┐
    │        Name          │ Version │                           Other Tags                             │
    ├──────────────────────┼─────────┼──────────────────────────────────────────────────────────────────┤
    │ cluster-profile-test │ 1.3.0   │ —                                                                │
    ├──────────────────────┼─────────┼──────────────────────────────────────────────────────────────────┤
    │ cluster-profile-test │ 1.2.0   │ —                                                                │
    ├──────────────────────┼─────────┼──────────────────────────────────────────────────────────────────┤
    │ cluster-profile-test │ 1.1.0   │ —                                                                │
    └──────────────────────┴─────────┴──────────────────────────────────────────────────────────────────┘

  3. Send a query asking for information about your Palette clusters. For example, you can ask "Which Palette clusters are deployed to AWS?" to list your AWS clusters.

    Example Output
    ⏺ palette - read_clusters (MCP)(filters: { cloud_type: "aws" })

    ⏺ You have 3 clusters deployed to AWS (cloud_type: aws):
    ┌───────────────┬─────────┐
    │      Name     │  State  │
    ├───────────────┼─────────┤
    │ prod-cluster  │ Running │
    ├───────────────┼─────────┤
    │ dev-cluster   │ Running │
    ├───────────────┼─────────┤
    │ cluster-test  │ Running │
    └───────────────┴─────────┘

    The remaining clusters run on vSphere and MAAS.

Read Cluster Status​

The read_cluster_status tool returns the state, conditions, and other status details for a single cluster. It is the recommended polling tool after a create_cluster or delete_cluster call, because a state-only projection costs a single API call.

  1. Open your MCP client.

  2. Send a query asking for the status of a cluster. For example, you can ask "What is the current status of the <cluster-name> cluster?".

    Replace the placeholder with your cluster name.

    Example Output
    ⏺ palette - read_cluster_status (MCP)(uid: "650f1a2b3c4d5e6f7a8b9c0d", fields: ["status.state"])

    ⏺ The prod-cluster cluster is currently in the Running state.

Read Cluster Observability​

The read_cluster_observability tool returns compliance-scan, backup, and restore results for a single cluster.

  1. Open your MCP client.

  2. Send a query asking for observability data for a cluster. For example, you can ask "Show me the latest compliance scans and backup status for the <cluster-name> cluster."

    Replace the placeholder with your cluster name.

    Example Output
    ⏺ palette - read_cluster_observability (MCP)(uid: "650f1a2b3c4d5e6f7a8b9c0d", include: ["scans", "backup"])

    ⏺ Compliance scans and backup status for prod-cluster:
    - Latest CIS scan: 2 High, 5 Medium findings
    - Latest backup: succeeded 2 hours ago

Retrieve a Cluster kubeconfig​

The read_cluster_kubeconfig tool retrieves the kubeconfig file for a cluster. The tool supports read-only, admin, and OIDC modes. To learn more about kubeconfig access permissions in Palette, refer to kubeconfig Access Permissions.

  1. Open your MCP client.

  2. Send a query asking to retrieve the kubeconfig for a cluster. For example, you can ask "Download the kubeconfig for the <cluster-name> cluster."

    Replace the placeholder with your cluster name.

    Example Output
    ⏺ palette - read_cluster_kubeconfig (MCP)(uid: "650f1a2b3c4d5e6f7a8b9c0d")

    ⏺ The kubeconfig for prod-cluster has been retrieved. Save it locally and set the KUBECONFIG environment variable to use it with kubectl.

Manage Cluster Tags​

The read_cluster_tags tool returns the tag set on a cluster. The update_cluster_tags tool replaces a cluster's tag set. Tag management through the MCP server is available for clusters only. Tag management for cluster profiles, cluster templates, edge hosts, and cluster template policies is not supported.

  1. Open your MCP client.

  2. Send a query asking to view the tags on a cluster. For example, you can ask "What tags are set on the <cluster-name> cluster?".

    Replace the placeholder with your cluster name.

    Example Output
    ⏺ palette - read_cluster_tags (MCP)(uid: "650f1a2b3c4d5e6f7a8b9c0d")

    ⏺ The prod-cluster cluster has the following tags:
    - env:prod
    - owner:platform-team

  3. Send a query asking to update the tags on a cluster. For example, you can ask "Add the cloud:aws tag to the <cluster-name> cluster."

    Replace the placeholder with your cluster name.

    Example Output
    ⏺ palette - update_cluster_tags (MCP)(uid: "650f1a2b3c4d5e6f7a8b9c0d", tags: ["env:prod", "owner:platform-team", "cloud:aws"])

    ⏺ Done. The prod-cluster cluster tags are now:
    - env:prod
    - owner:platform-team
    - cloud:aws (new)

Read Component Events​

The read_events tool returns paginated Kubernetes and controller component events. You can scope the query to one cluster, one edge host, or a recent tenant-wide feed.

  1. Open your MCP client.

  2. Send a query asking for events. For example, you can ask "Show me the last 20 error events for the <cluster-name> cluster."

    Replace the placeholder with your cluster name.

    Example Output
    ⏺ palette - read_events (MCP)(object_kind: "spectrocluster", object_uid: "650f...", limit: 20, filters: { severity: "Error" })

    ⏺ Latest error events for dev-cluster-1:
    - FailedScheduling — 0/3 nodes available: insufficient memory (2 minutes ago)

Manage Projects, Teams, and Users​

The read_projects, read_teams, and read_users tools return project, team, and user records. The corresponding create_*, update_*, and delete_* tools mutate those records when the server runs with --allow-write.

  1. Open your MCP client.

  2. Send a query asking to create a project. For example, you can ask "Create a Palette project named <project-name>."

    Replace the placeholder with your cluster name.

    Example Output
    ⏺ palette - create_project (MCP)(name: "dev-project")

    ⏺ Done. The dev-project project has been created with uid 650f1a2b3c4d5e6f7a8b9c0d.

  3. Send a query asking to list team membership. For example, you can ask "List the members of the <team-name> team."

    Replace the placeholder with your team name.

    Example Output
    ⏺ palette - read_teams (MCP)(uid: "qa-team")

    ⏺ The qa-team team has 2 members: alice@example.com, bob@example.com.

Delete Clusters and Cluster Profiles​

The delete_cluster tool deletes a cluster. The delete_cluster_profile tool deletes a cluster profile. Both tools require a typed-name confirmation before the delete is issued to Palette.

  1. Start the Palette MCP server with the --allow-write startup flag to enable write tools. Refer to the applicable setup guide for Claude, Cursor, or Antigravity for more information.

  2. Open your MCP client.

  3. Send a query to delete a Palette cluster. For example, you can say "Delete the <cluster-name> Palette cluster."

    Replace the placeholder with your cluster name.

    The MCP server prompts you to type back the cluster name to confirm the delete.

    Example Output
    ⏺ This will delete the running dev-cluster cluster (uid: 69c401bf35d56c1812c5a00f). This is irreversible.
      To confirm, please type the cluster name: dev-cluster

    ❯ dev-cluster

    ⏺ palette - delete_cluster (MCP)(uid: "69c401bf35d56c1812c5a00f", confirm: "dev-cluster")

    ⏺ Done. The dev-cluster cluster has been marked for deletion.

  4. Poll the cluster status with read_cluster_status until the cluster reaches the Deleted state.

    Example Output
    ⏺ palette - read_cluster_status (MCP)(uid: "69c401bf35d56c1812c5a00f", fields: ["status.state"])

    ⏺ The dev-cluster cluster is currently in the Deleting state.

  5. Send a query to delete a Palette cluster profile. For example, you can say "Delete the <cluster-profile-name> Palette cluster profile."

    Replace the placeholder with your cluster profile name.

    Example Output
    ⏺ This will delete the cluster-profile-test cluster profile (uid: 698bc897c7fdfaca31d7122f). This is irreversible.
      To confirm, please type the cluster profile name: cluster-profile-test

    ❯ cluster-profile-test

    ⏺ palette - delete_cluster_profile (MCP)(uid: "698bc897c7fdfaca31d7122f", confirm: "cluster-profile-test")

    ⏺ Done. The cluster-profile-test cluster profile has been deleted.