Deployer health check blocks deploys to remote/ephemeral instances (requires a local gateway)

Summary

The localstack-deployer tool runs a pre-deploy reachability check against a local gateway (http://${LOCALSTACK_HOSTNAME:-localhost}:${LOCALSTACK_PORT:-4566}/_localstack/health). When you want the deploy to target a remote Ephemeral Instance (which is HTTPS-only, e.g. https://ls-<id>.sandbox.localstack.cloud), that check can never pass on its own, so the deploy is blocked with:

LocalStack Not Running — LocalStack is not reachable at http://localhost:4566 ...

even though the actual deploy would succeed against the ephemeral instance. The result is that you must keep a throwaway local LocalStack running purely to satisfy the health gate, while the real deploy is redirected elsewhere.

Why this matters

A natural workflow is to give each agent (or each CI job / worktree) its own Ephemeral Instance in the cloud and deploy straight to it via the MCP server. The deployer can already do this: tflocal/cdklocal pick up the endpoint from the environment, and setting per-service {SERVICE}_ENDPOINT=https://ls-<id>.sandbox.localstack.cloud in the MCP server's env makes terraform apply land resources on the ephemeral instance correctly. The only thing standing in the way is the pre-deploy health probe.

Details

• The env builder forwards the process env to tflocal, so per-service *_ENDPOINT https values reach the provider and resources deploy to the ephemeral instance as expected.
• The health gate, however, is hardcoded to an http://host:port probe. Ephemeral instances only answer on https://<host> (443); http://<host>:443 and http://<host>:80 do not respond, so the gate fails.
• Confirmed: with a local LocalStack running (gate satisfied) + per-service *_ENDPOINT=https://ls-<id>... set, a Terraform deploy via localstack-deployer lands entirely on the ephemeral instance. With the local instance stopped, the same deploy is refused by the gate before tflocal ever runs.

Steps to reproduce

1. Create an ephemeral instance: localstack ephemeral create --name demo -> note the https://ls-<id>.sandbox.localstack.cloud endpoint.
2. Configure the MCP server env with AWS_ENDPOINT_URL and per-service {SERVICE}_ENDPOINT all set to that https endpoint (and the auth token).
3. Stop any local LocalStack.
4. Call localstack-deployer (deploy, terraform) on a small project.
5. It fails with "LocalStack is not reachable at http://localhost:4566" instead of deploying to the ephemeral instance.

Suggested fix

When a remote endpoint is configured (e.g. AWS_ENDPOINT_URL / *_ENDPOINT points at a non-local host, or an explicit ephemeral endpoint is provided), the pre-deploy reachability check should probe that endpoint (honouring https), or be skipped, rather than always probing http://localhost:4566. Ideally the tool would accept an explicit target/endpoint so a deploy can be aimed at an ephemeral instance without needing a local gateway at all.

Environment

• @localstack/localstack-mcp-server (latest via npx)
• terraform-local (tflocal) 0.26.0, Terraform 1.14.9
• Ephemeral Instances (public preview), image localstack/localstack-pro:latest

Beginne damit, den im Issue beschriebenen pre-deploy health check und den env builder von localstack-deployer zu finden. Reproduziere den Fehler mit einem entfernten HTTPS-ephemeren Endpoint und ohne lokales Gateway. Überprüfe anschließend, dass der konfigurierte entfernte Endpoint geprüft wird oder der Check übersprungen wird und dass der Deploy über tflocal fortgesetzt wird.