Skip to content

Example prompts

wiki-sync edited this page Jul 19, 2026 · 1 revision

Example prompts

Copy-paste these into Cursor chat once the proxmox MCP server shows green (~155 tools). Not wired yet? Start with Setup.

Put API tokens in config.json / env — not in prompts.

Before you paste

  • Create ≠ ready: create tools return an async UPID. The agent should call wait_for_task before starting the guest or reporting success.
  • LXC create ≠ app deploy: create_lxc provisions an OS template only. Nesting features do not install Docker — see Recipes (Nested Docker).
  • Runtime IP / guest exec: execute_lxc_command and reliable runtime IPs need opt-in host SSH (Setup).

Starter prompts

1. First connection

Call get_nodes and get_version to confirm the Proxmox MCP connection is working. Summarize what you find.

2. Cluster overview

Using the Proxmox MCP tools, give me a quick health summary of my cluster — nodes, storage pools, and what's currently running.

3. Deploy this project in Docker

Analyze this workspace to determine the resource requirements and the best base OS. Then, use the Proxmox tools to spin up a new LXC container with Docker to host this project — prefer an Ubuntu template if one is available, otherwise choose the best match from list_os_templates. Let me know the IP address when it is ready.

What to expect: analyze repo → get_next_vmid / list_os_templatescreate_lxc with nesting=1,keyctl=1wait_for_taskstart_lxc → install Docker → report IP. For reliable IP + install, enable host SSH in config or pass ssh_public_keys on create. Deeper walkthrough: SETUP.md — Provision a nested Docker LXC.

4. Simple LXC (no Docker)

Create a small Ubuntu LXC container with 2 GB RAM on my cluster. Wait for creation to finish, start it, and tell me how to reach it.

5. Inventory

List all VMs and containers on the cluster. Include status and which node each guest is on.

6. Storage pressure

Which storage pools are low on space? List recent backups if you can.

7. Safe changes

Create a snapshot of guest 105 named before-changes before we modify anything.

8. Recent tasks

Show recent cluster tasks and highlight any failures.

9. Empty lists / permissions

My get_containers list is empty but I know containers exist. Check my API token permissions and tell me what ACL fixes I need.

10. VM from ISO

List available ISOs, create a new VM with an Ubuntu ISO and cloud-init SSH access, wait for creation to complete, and report the VM ID.


Development & DevOps

These prompts assume MCP is connected and you are working in a Cursor workspace. They combine repo context with Proxmox ops — useful for local dev, CI, and deploy workflows.

Cursor tip: Open the repo you want deployed and @-mention key files (Dockerfile, compose.yaml, package.json) in the same chat so the agent can size CPU/RAM/ports correctly.

Ephemeral dev environment from this repo

Use when: you want a throwaway guest sized for the current project.

Analyze this workspace (runtime, ports, env vars, dependencies). Provision a throwaway LXC or VM on Proxmox sized for local development, install what's needed, and give me SSH/IP details to deploy this project. Name it with a dev- prefix so we can clean it up later.

Snapshot before risky change

Use when: you are about to change a guest and want a rollback point.

I'm about to change guest 105. Create a snapshot named pre-deploy-$(date) (or today's date), confirm it succeeded, then proceed only after the snapshot task completes.

Post-config verification

Use when: you edited guest config and need to know if a reboot is required.

I just updated the config on guest 120. Check get_guest_pending, tell me if a reboot is required, and reboot only if pending changes block what we need.

Debug a failing guest

Use when: something is wrong and you want diagnosis before any restart or delete.

Guest 108 isn't behaving. Pull its status, recent cluster tasks for that guest, and network info. Summarize likely causes and suggest the safest next step — don't restart or delete without asking.

Backup before release

Use when: you want a backup volume before deploying.

Create a backup of guest 105 to my default backup storage before we deploy. Wait for the task to finish and report the backup ID/volume name.

Tear down ephemeral dev resources

Use when: cleaning up dev-* guests after experiments.

List containers and VMs with names starting with dev-. For any that are stopped and older than a week (or that I confirm), delete them with force=true only after I approve each one.

Multi-server cluster

Use when: you need several guests for a distributed app, k8s lab, database replica set, etc.

I need 3 Ubuntu servers for a [describe purpose, e.g. Redis cluster / k8s control plane + workers]. Before creating anything: check cluster capacity (get_cluster_resources, storage, node memory), propose VM vs LXC and sizing per node, and pick a consistent naming scheme (redis-01, …). Provision guests one at a timewait_for_task after each create before starting the next. Spread across Proxmox nodes when possible. When all are up, give me a summary table: name, VMID, node, IP, role. Ask before deletes or force operations.

Docker Swarm with scalability

Use when: you want orchestration across multiple Proxmox guests, not just a single Docker host.

Design and provision a Docker Swarm lab on Proxmox: 1 manager + 2 workers (Ubuntu LXC with Docker unless you recommend VMs). Analyze workspace/cluster capacity first, use nesting features where needed, install Docker on each host, initialize Swarm on the manager, join the workers, and verify with docker node ls. Scale pattern: name guests swarm-mgr-01, swarm-wkr-01, swarm-wkr-02 so we can add swarm-wkr-03 later. Report a host table (name, VMID, node, IP, role) when ready. Do not claim the stack is production-ready until nodes are healthy and joined.

Multi-guest tips

  • Say how many nodes and what roles up front; vague “spin up a cluster” leads to wrong sizing.
  • Agents should serialize creates (wait_for_task per guest) — parallel creates race VMIDs and obscure failures.
  • For Swarm/k8s, mention whether you need nested LXC or full VMs; see Recipes for LXC nesting limits.

Write better prompts

  • Be explicit about wait-for-ready, desired node/bridge if you care, and “don’t claim the app is live until a port responds.”
  • Not everything the model suggests is available via the Proxmox API (host package upgrades are a common gap). Prefer prompts that match registered tools — see API coverage.

See also

Recipes · Tools · Troubleshooting · SETUP.md — example prompts