Skip to main content
Glama

Create AWS cluster

create_aws_cluster
Destructive

Create and provision a new self-managed Kubernetes cluster on AWS EC2: k3s or kubeadm on Ubuntu instances reached through a bastion host, with the AWS cloud controller manager, the EBS CSI driver and an NLB-fronted ingress installed. By default Ankra creates the network: a VPC from network_ip_range (default 10.0.0.0/16) with a public /24 and a private /20 per availability zone, an internet gateway and NAT gateways (egress_mode nat_gateway), all torn down with the cluster; omit vpc_id for this and, unless the user names availability_zones, the preflight picks one zone (three when control_plane_count is 3 or more). To adopt an existing VPC instead, pass vpc_id with at least one private node subnet (node_subnet_ids) and a public bastion subnet (bastion_subnet_id); Ankra never modifies an adopted VPC's network. Requires an AWS credential, an SSH key credential, a region and the IPv4 CIDRs allowed to SSH to the bastion. Use list_provider_credentials with provider='aws' and provider='ssh-key' to find the credential ids. There is no chat-side VPC or subnet listing: when adopting, take the vpc_id, node_subnet_ids and bastion_subnet_id from the user (or the AWS console) and confirm with them, before proposing, that every subnet belongs to that VPC and region, the node subnets are private, and the bastion subnet routes to an internet gateway - the same checks the create's preflight enforces. The create runs that preflight first (credential scope; for a created network the CIDR plan, the zones and the VPC / NAT-gateway / elastic-IP quotas, for an adopted VPC the VPC and subnet facts and egress routes; instance-type availability, service quotas) and refuses with every failing item when it cannot proceed; nothing is written on refusal. Instance types are EC2 names (e.g. 't3.medium', 'm6i.large'). Managed EKS creation is create_managed_cluster, not this tool. Only available to organisations with AWS self-managed provisioning enabled.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "credential_id": "<credential_id>",
      "ssh_key_credential_id": "<ssh_key_credential_id>",
      "region": "<region>",
      "bastion_allowed_ips": "<bastion_allowed_ips>",
      "name": "<name>"
    },
    "name": "create_aws_cluster"
  }
}

Input Schema

NameRequiredDescriptionDefault
cniNoContainer network plugin. Default: 'cilium'. Cilium and Calico enforce the IMDS guard (the network policy that keeps pods off the instance metadata service and the node's instance role); flannel cannot enforce it, so a flannel cluster is accepted with a warning that pods on control-plane nodes can reach the control-plane instance role.
nameYesCluster name. Must be unique within the organisation.
regionYesAWS region the cluster (and its VPC) lives in (e.g., 'eu-west-1', 'us-east-1').
vpc_idNoId of an existing VPC to adopt (e.g., 'vpc-0abc...'); node_subnet_ids and bastion_subnet_id are then required and network_ip_range, availability_zones and nat_gateway_single_zone are refused. Omit it to have Ankra create the VPC and its subnets.
descriptionNoOptional cluster description.
egress_modeNoHow the private node subnets reach the internet. Created VPC: 'nat_gateway' (default; Ankra-created NAT gateways with their own elastic IPs) or 'bastion_nat' (the bastion is the NAT instance); 'existing' is refused. Adopted VPC: 'existing' uses the subnets' own NAT gateway or egress route, 'bastion_nat' makes the bastion the NAT for them, 'nat_gateway' is refused (it would re-point the customer's route tables); omit to let the preflight resolve it from the subnets' route tables.
environmentNoWhat the cluster is for. Ankra's autonomy gates treat a cluster like production until a person confirms what it is for. A label named here is that confirmation only when a member confirms this call's card in Ankra chat; over MCP, or with no member's confirmation, it is recorded as set by automation and confirms nothing (a production label is still honoured). Omit it and the cluster is labelled development, recorded as a default nobody confirmed.
gateway_apiNoAlso serve the Kubernetes Gateway API: the cluster's Traefik serves Gateway API routes (HTTPRoute) next to Ingress, and Ingress keeps working. Adds the Gateway API CRDs, a shared traefik-gateway Gateway in the traefik namespace (a plain HTTP listener that accepts routes from every namespace), and cert-manager and external-dns support for those routes. Needs the networking stack (on by default): asking for it with include_networking=false is refused. Default: false. Set at create only; no tool turns it on or off on an existing cluster.
include_dnsNoProvision the cluster's Ankra-managed DNS zone. Default: true.
worker_typeNoEC2 instance type for worker nodes. Default: 't3.medium'.
distributionNoKubernetes distribution. Default: 'kubeadm' (vanilla upstream Kubernetes with Cilium). 'k3s' is the lightweight alternative. kubeadm supports only the cilium CNI on AWS.
worker_countNoNumber of worker nodes. Default: 1.
credential_idYesUUID of the AWS credential. Get from list_provider_credentials with provider='aws'. A role credential must have been onboarded with the 'provisioning' or 'self_managed' scope; an access-keys credential is accepted and the preflight probes decide.
gitops_branchNoBranch of the GitOps repository. Default: 'master'.
ubuntu_seriesNoUbuntu LTS series for the node image, resolved from Canonical's public AMI catalogue (e.g., '24.04'). Default: '24.04'.
gitops_providerNoProvider of the GitOps repository: 'github' (the default when omitted) or 'bitbucket_cloud'. Pass the credential's provider from list_gitops_credentials.
node_subnet_idsNoAdopted VPC only: private subnet ids (e.g., 'subnet-0abc...') in the VPC for the control-plane and worker nodes; at least one, spread across availability zones for HA. They must not route to an internet gateway. Refused without vpc_id - a created VPC lays out its own subnets.
root_volume_gibNoRoot volume size in GiB for every node (gp3). Default: 40.
gitops_repo_slugNoBitbucket Cloud repository slug of the GitOps repository (from list_gitops_repositories). Only with gitops_provider 'bitbucket_cloud', together with gitops_workspace.
gitops_workspaceNoBitbucket Cloud workspace of the GitOps repository (from list_gitops_repositories). Only with gitops_provider 'bitbucket_cloud', together with gitops_repo_slug.
network_ip_rangeNoCreated VPC only: the IPv4 CIDR the VPC is created with, /16 to /20 as a network address (default '10.0.0.0/16'); it must hold a private /20 and a public /24 per availability zone. Refused alongside vpc_id.
retention_policyNoStorage retention on teardown. Default: 'retain'.
bastion_subnet_idNoAdopted VPC only: public subnet id in the VPC for the bastion host; it must route to an internet gateway so the bastion gets a reachable public address. Refused without vpc_id.
gitops_repositoryNoOptional GitOps repository to connect at creation: 'owner/repo' on GitHub, 'workspace/repo_slug' on Bitbucket Cloud; requires gitops_credential_name.
availability_zonesNoCreated VPC only: the zones the VPC is laid out over, in order (the bastion and the first control plane land in the first), e.g. ['eu-west-1a', 'eu-west-1b', 'eu-west-1c']. More than one zone needs control_plane_count >= 3. Omit to let the preflight pick one zone (three when control_plane_count >= 3) from the region's available zones. Refused alongside vpc_id.
control_plane_typeNoEC2 instance type for control plane nodes. Default: 't3.medium'.
include_networkingNoInstall the platform networking stack (ingress, cert-manager, external-dns). Default: true.
kubernetes_versionNoKubernetes version. If not specified, uses the latest supported version.
bastion_allowed_ipsYesIPv4 CIDRs allowed to SSH to the bastion (e.g., '203.0.113.4/32'); IPv6 is accepted only as the open-world '::/0', and any other IPv6 CIDR is refused. Required: name at least one, or the call is refused. '0.0.0.0/0' (and '::/0') is accepted rather than refused - it leaves SSH reachable from the whole internet, where access is still key-only with password login off and sshd rate-limited on the host. Preflight reports the exposure either way in bastion_ssh_exposure: a warning when the list is open, ok when it is narrowed. Name the operator's own CIDRs to restrict it.
control_plane_countNoNumber of control plane nodes. Default: 1.
bastion_instance_typeNoEC2 instance type for the bastion host. Default: 't3.small'.
ssh_key_credential_idYesUUID of the SSH key credential. Get from list_provider_credentials with provider='ssh-key'.
gitops_credential_nameNoName of the git credential, GitHub or Bitbucket Cloud, that can read the GitOps repository (see list_gitops_credentials). Requires gitops_repository, or for Bitbucket Cloud gitops_workspace and gitops_repo_slug.
nat_gateway_single_zoneNoCreated VPC with nat_gateway egress only: true creates one NAT gateway in the first zone that every private subnet routes through (cheaper, one egress failure domain) instead of one per zone. Default: false. Refused when true alongside vpc_id.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changed
    • addedInput schema / properties / environment
      Added value: +{
      +  "description": "What the cluster is for. Ankra's autonomy gates treat a cluster like production until a person confirms what it is for. A label named here is that confirmation only when a member confirms this call's card in Ankra chat; over MCP, or with no member's confirmation, it is recorded as set by automation and confirms nothing (a production label is still honoured). Omit it and the cluster is labelled development, recorded as a default nobody confirmed.",
      +  "enum": [
      +    "development",
      +    "staging",
      +    "testing",
      +    "production",
      +    "local"
      +  ],
      +  "type": "string"
      +}
  2. Changed1 schema field changed
    • addedInput schema / properties / gateway_api
      Added value: +{
      +  "description": "Also serve the Kubernetes Gateway API: the cluster's Traefik serves Gateway API routes (HTTPRoute) next to Ingress, and Ingress keeps working. Adds the Gateway API CRDs, a shared traefik-gateway Gateway in the traefik namespace (a plain HTTP listener that accepts routes from every namespace), and cert-manager and external-dns support for those routes. Needs the networking stack (on by default): asking for it with include_networking=false is refused. Default: false. Set at create only; no tool turns it on or off on an existing cluster.",
      +  "type": "boolean"
      +}
  3. First observed

TDQS

A4.2/5.0
Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already flag destructiveHint=true and non-idempotent, and the description adds real value on top: a preflight that refuses with every failing item and writes nothing on refusal, network teardown with the cluster, no modification of an adopted VPC, and an organisation-level availability gate. It does not, however, say what the call returns or whether creation is asynchronous (long-running with a follow-up execution), which matters for a destructive multi-node provisioning tool.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the action and scope, and essentially every sentence carries operational information (credit paths, refusals, subnet requirements). The weakness is structure: it is one long undifferentiated paragraph with heavily nested clauses, when a 34-parameter destructive tool would benefit from bullets or short labelled blocks.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness3/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Prerequisites, preflight behavior, network topology, and credential sourcing are covered thoroughly for a complex 34-parameter tool. Since there is no output schema, the exclusion of return shape, timing, and whether an execution must be awaited with wait_for_execution leaves a gap that matters for an agent coordinating the call.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100% across 34 parameters, so the schema does the heavy lifting; the description mostly restates defaults. It does add relational meaning the schema only partially carries, notably the CIDR layout derived from network_ip_range, the zone-selection rule tied to control_plane_count, and how egress_mode is resolved from route tables for an adopted VPC.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

Opens with a concrete verb and resource ('Create and provision a new self-managed Kubernetes cluster on AWS EC2') and then specifies the distribution, OS, access path, and installed components. It explicitly rules out the nearest sibling ('Managed EKS creation is create_managed_cluster, not this tool'), so an agent can separate it from create_managed_cluster or create_ankracloud_cluster without opening a schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives explicit when-to-use branches: create-the-network by default versus adopt-an-existing-VPC by passing vpc_id with node_subnet_ids and bastion_subnet_id, plus the checks to confirm with the user before proposing. It also names prerequisites (AWS and SSH-key credentials, region, bastion CIDRs) and points at list_provider_credentials and create_managed_cluster as alternatives.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.