happypod-mcp
OfficialServer Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| HAPPYPOD_MODE | No | Set to `demo` to use the built-in simulator. No account or billing is needed in demo mode. | |
| HAPPYPOD_API_KEY | No | Your API key from happypod.io → Settings → API keys. | |
| HAPPYPOD_API_URL | No | Optional, another API base URL. Default https://api.happypod.io/v1. | https://api.happypod.io/v1 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Happy Pod is the DevOps AI agent for vibe coders: it ships, scales, watches and fixes apps, GPU pods and managed databases. Rules: (1) before anything that costs money (create_pod, create_gpu_pod, scale_pod, paid create_database) call it with dryRun: true, tell the user the price, and wait for a clear yes; (2) before delete_* say exactly what will be lost and wait for a yes, then pass the exact name as confirm; (3) deployments are asynchronous: poll get_deployment until success or failed, and on failed read get_deployment_logs and explain the cause; (4) never print secret values or revealed connection URLs; (5) when something is broken or slow, start with diagnose_pod.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| whoamiA | The account behind the API key: name, plan, credit balance (USD), spend this month and the key's scope. Call it first in a session. mode "demo" means the built-in simulator, not a real account. |
| list_regionsA | Regions where pods and databases can run, with their city and whether GPUs are available there. Pick the one closest to the app's users. |
| list_sizesA | CPU pod sizes (vCPU, memory, hourly and monthly USD price). Start small: most vibe-coded apps run on |
| list_gpu_typesA | GPU types for GPU pods: VRAM, hourly USD price and the regions with free capacity right now. |
| list_projectsA | Projects this key can reach, each with its pod and database counts. Start here to find a projectId. |
| create_projectB | Create an empty project (free). One project per app is a good default. |
| delete_projectA | Delete an EMPTY project. Irreversible. Refused with 409 PROJECT_NOT_EMPTY while pods or databases remain. Only after the user agrees; confirm must be the project's exact name. |
| list_podsB | Pods in a project: app pods (web apps, APIs, workers) and GPU pods, with status, size, replicas, URL and the last deployment. |
| get_podA | One pod in full: source, size, replicas, region, ports, URL, domains, status and the last deployment. |
| create_podA | Create an app pod and start its first deployment. Source is a Git repository (built automatically: a Dockerfile is used when present, otherwise the stack is detected) or a container image. ALWAYS call with dryRun: true first, tell the user the price, and create only after a clear yes. Then poll get_deployment until success or failed. |
| create_gpu_podA | Start a GPU pod from a container image (inference servers, notebooks, fine-tuning, batch jobs). Billed per hour while running. ALWAYS dryRun first and tell the user the hourly price; stop it with stop_pod when the job is done. |
| deploy_podA | Build and roll out the pod again from its source (latest commit of the branch, or the image tag). Needed after changing variables. Zero-downtime: the old version serves until the new one is healthy. |
| list_deploymentsA | A pod's deployments, newest first: id, status, commit or image, start and finish time. Use it to pick a deploymentId for rollback_pod. |
| get_deploymentA | One deployment's status: queued, building, deploying, success or failed. Poll every few seconds after create_pod or deploy_pod; builds take one to five minutes. |
| get_deployment_logsA | Build and start-up log of a deployment. Read it when a deployment failed: the cause is usually a missing variable, the wrong port or a build error. |
| get_pod_logsB | Runtime logs of a running pod across replicas. |
| get_pod_metricsB | CPU, memory, requests per second, error rate and p95 latency over a window, per replica and in total. |
| diagnose_podA | Happy Pod's own health check for a pod: reads the last deployment, recent error logs and metrics, and returns findings (severity, evidence, suggested fix). Start here when the user says something is broken or slow. |
| restart_podA | Restart the running replicas without rebuilding. Brief interruption per replica. |
| stop_podA | Stop a pod (billing for compute stops; volumes are kept). GPU pods should be stopped as soon as their job is done. |
| start_podA | Start a stopped pod again. Billing resumes. |
| scale_podA | Change a pod's size and/or replica count. dryRun: true first, show the user the current and new monthly price, apply after a clear yes. |
| rollback_podA | Put an earlier successful deployment back live (no rebuild). The quickest fix when a new deploy broke the app. |
| delete_podA | Delete a pod, its deployments, domains and volumes. Irreversible. Only after the user agrees; confirm must be the pod's exact name. |
| list_envB | A pod's environment variables. Secret values come back masked. |
| set_envA | Create or update variables. Put credentials in secrets. Changes apply on the next deploy: call deploy_pod afterwards. |
| unset_envB | Remove one variable. Applies on the next deploy. |
| add_domainA | Attach a custom domain to a pod's public port. Returns the DNS record the user must create (CNAME for subdomains, A/AAAA for an apex). HTTPS is issued automatically once DNS points here. |
| list_domainsB | A pod's domains with DNS and certificate status (pending_dns, issuing, active). |
| list_databasesC | Managed databases in a project: engine, plan, status, region. |
| create_databaseA | Create a managed Postgres or Redis database with daily backups. The hobby plan is free. dryRun first for paid plans. Then pass its connection URL to the app as a secret (get_database_connection + set_env). |
| get_database_connectionA | Host, port, database and user, plus the connection URL with the password masked. reveal: true returns the real URL: use it only to pass into set_env secrets, never print it to the user. |
| delete_databaseA | Delete a database and all its data and backups. Irreversible. Only after the user agrees; confirm must be the database's exact name. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 33 tools
Most tools map cleanly to a unique resource and action, with strong descriptions explaining adjacent pod lifecycle operations. A few could be confused at a glance—start_pod/restart_pod/deploy_pod/rollback_pod and create_pod/create_gpu_pod—but the descriptions usually disambiguate.
The set follows a very consistent snake_case verb_noun pattern: list_pods, get_pod, create_database, delete_project, set_env, etc. whoami is the only exception, but it is a conventional standalone command and does not undermine the pattern.
33 tools is heavy for an agent-facing MCP surface, even for a broad PaaS. Most tools are genuinely distinct, but several pod lifecycle actions could potentially be consolidated, making this borderline rather than clearly excessive.
The core deployment lifecycle is well covered: projects, pods, GPU pods, env vars, domains, databases, deployments, logs, metrics, and diagnostics. Notable gaps remain around domain deletion, database resize/backup-restore, and deployment cancellation, but core workflows are usable.