v0.3.0-rc.3

Release v0.3.0-rc.3

v0.3.0-rc.2

Release v0.3.0-rc.2

v0.3.0-rc.1

Release v0.3.0-rc.1

v0.2.0

Added

• Instances: rebuild_instance ★ and restore_instance_snapshot ★ (destructive: the preview says
  what data is lost and that the AI app needs "Manage and delete" access), get_instance_metrics,
  get_instance_usage, list_instance_volumes, list_instance_snapshots,
  create_instance_snapshot ★ (priced from the snapshot quote) and get_instance_order (with
  wait to follow the order and its create operation).
• Temporary SSH access: open_ssh_access ★ opens port 22 for 15 minutes to the address Pantech
  sees the request come from. The preview names that address as this server sees it (the caller's
  address behind a trusted proxy; "this computer's public address" in stdio mode) and warns that
  through a web-based AI app it is the app's server, not the user. Once confirmed it waits for the
  grant and returns the host, port, expiry and a ready-to-run ssh command; the user's own SSH key
  is never seen. get_ssh_access, and revoke_ssh_access (one step: it only closes the port).
• Networking: delete_network ★, delete_subnet ★, delete_public_ip ★,
  list_port_forwarding_rules, delete_port_forwarding_rule ★, update_load_balancer ★
  (replacing targets is confirmed), delete_load_balancer ★, create_security_group,
  delete_security_group ★.
• Storage: delete_snapshot ★, restore_snapshot ★ (a new, billed volume, priced from the disk
  offering quote), get_snapshot_schedule, set_snapshot_schedule ★ (the preview says when
  lowering retention deletes old snapshots), get_volume_usage.
• Databases: start_database, stop_database, resize_database_storage ★ (priced from the
  engine's storage price), delete_database ★, list_database_snapshots,
  create_database_snapshot ★, get_database_order.
• Kubernetes: delete_kubernetes_cluster ★, start_kubernetes_cluster, stop_kubernetes_cluster,
  list_node_pools, create_node_pool ★, delete_node_pool ★, list_kubernetes_addons,
  toggle_kubernetes_addon ★, list_cluster_nodes, list_cluster_workloads.
• SSH keys: delete_ssh_key ★.
• Billing and webhooks: get_usage_by_resource, list_receipts, list_webhook_deliveries (read
  only; endpoint responses are not returned).
• Every tool that calls a route the API marks destructive is checked to be two-step and to say it
  needs "Manage and delete" access. Kubernetes manifests (read or applied) are added to the routes no
  tool may call, since they can hold Kubernetes secrets.

Changed

• get_credit_balance states the available credit as an exact figure and labels the next invoice
  as a forecast that grows with usage until the bill date, for example: "Available credit:
  NGN 12,345.67 (exact, as of now). Next invoice so far: NGN 1,234.00 on 2026-11-01 — a forecast
  that grows with usage until the bill date." Its description tells the app to report amounts
  exactly, never rounded.
• estimate_price says its total is an estimate; get_usage and the usage tools say their figures
  are exact measured quantities. Money is formatted with integer arithmetic in each currency's minor
  unit (kobo, cents), so no amount is rounded.
• Temporary SSH access is no longer on the excluded list: a grant holds no secret.

v0.1.0

First release.

• pantech-mcp stdio for a local AI app, with the Pantech CLI's sign-in or PANTECH_API_KEY.
• pantech-mcp http, the hosted Streamable HTTP server at https://mcp.pantechdynamics.com/mcp:
  OAuth through the console (RFC 9728 protected resource metadata), each caller's own key.
• 54 tools across instances, SSH keys, VPC networks, subnets and firewalls, public IPs and port
  forwarding, security groups, load balancers, volumes, snapshots, managed databases, Kubernetes,
  billing, usage, webhooks and operations.
• Two-step confirmation (preview with the price, then a signed, 5-minute, argument-bound token)
  for everything that costs money, opens access from the internet, or deletes or overwrites.
• The agent brief resource and the deploy_web_app, explain_my_bill and lock_down_firewall
  prompts.
• The Pantech logo in the server's identity, served at /icon.svg.
• pantech-mcp http checks each key with the API (GET /me) before serving /mcp, remembering
  only the answer by the key's hash (30 s for a good key, never past its expiry; 10 s for a bad
  one). An expired, revoked or unknown key gets 401 with
  WWW-Authenticate: Bearer error="invalid_token", resource_metadata="…", so AI apps sign in
  again; when the API cannot answer, 503 with Retry-After, never 401.
• Behind a trusted proxy (MCP_TRUST_PROXY=true), every API call carries the caller's address in
  X-Pantech-Client-IP, so the API's per-address limits and key audit see the AI app's address,
  not the MCP server's.
• Rate limits sized for hosted AI apps that share a few addresses: MCP_RATE_LIMIT_IP 1200 a
  minute, MCP_RATE_LIMIT_TOKEN 120, and a new MCP_RATE_LIMIT_AUTH_FAILURES (60 a minute per
  address) for requests without a good key.
• pantech-mcp stdio refuses a CLI config.json or credentials.json other users can write, and
  an API URL that is not https (or localhost); pantech-mcp http refuses an empty
  MCP_AUTHORIZATION_SERVER instead of falling back to the production console.