Skip to main content

Deploying

A deploy uploads a folder, builds it, starts it, checks that it's healthy and switches production to it. The console, the tc CLI and the MCP deploy tool all do the same thing through the same API.

Three ways to deploy​

FromHowGuide
The consoleProjects › New project, drop a folder; or Deploy a new version on a project's OverviewDeploy from the console
The terminaltc deploy [dir]Deploy with the CLI
An AI assistantThe MCP tool deploy (files or a tarball)Deploy with MCP

The CLI and MCP create the project on the first deploy if it doesn't exist.

What happens during a deploy​

StatusWhat vastframe is doing
queuedThe source bundle is stored and a deployment record exists
buildingDetecting the runtime, applying db/migrations/*.sql, creating declared buckets, generating a Dockerfile if needed, building the image (or publishing the static output)
deployingStarting the container and waiting for it to be healthy (container services only; static sites go straight to ready)
readyThe deployment serves its preview URL. vastframe then builds any cron and worker services, scans the code for integration use, files access requests and, unless it's a preview-only deploy, promotes it to production
failedSomething went wrong; the build log says what. Production keeps serving the previous deployment

Status streams live to the console and the CLI (server-sent events), and MCP's get_status returns it.

Step by step​

  1. Upload. The source arrives as a gzipped tarball. The CLI excludes node_modules, .git, .venv, venv, __pycache__, .env and .DS_Store; the console skips node_modules, .git, dist and .env files. Bundles are limited to 50 MB.
  2. Detect. vastframe reads tc.json if present, then looks for a Dockerfile, a package.json, Python requirements, or an index.html. See Runtimes and detection.
  3. Migrate. Each db/migrations/*.sql file runs once, in name order, as the database owner, before the new version is built and started. A failing migration fails the deploy and production is unchanged. Buckets declared in tc.json are created next. See Migrations.
  4. Build. Container services become an image; static sites become a folder of files that the Edge serves. If the app depends on the SDK, the build supplies the platform's own copy.
  5. Start and health check. The container starts with the platform environment and your variables, and must answer on $PORT within the health-check timeout (90 seconds by default).
  6. Cron and worker services. Each one is built (or reuses the primary image) and registered with its command and schedule.
  7. Scan for integrations. The code is scanned for SDK calls such as tc.int.salesforce.get(...), tc.m365.excel(...) and tc.teams.notify(...), with file and line. Together with tc.json integrations, this decides which grants the app needs. Anything not already granted or auto-approved becomes an access request.
  8. Promote. Production switches to the new deployment. The old one keeps its preview URL, so you can roll back instantly.

Deploys never fail for permissions​

If the app needs an integration scope that isn't granted yet, the deploy still completes. The app goes live, and the affected calls answer ACCESS_PENDING with the request number and approvers until someone approves. Then they start working, with no redeploy. The CLI prints the requests a deploy created:

Live: https://escalations.apps.contoso.com
Preview: https://escalations--k2m9qa.apps.contoso.com
Access request #42: salesforce Case:read, Account:read → waiting for Sam Taylor (calls answer ACCESS_PENDING until approved)

When an agent deploys over MCP and something needs approval, deploy requires a one-line purpose that approvers will read.

Redeploying without a rebuild​

Changing environment variables doesn't rebuild anything. Redeploy (Redeploy to apply on the project's Settings page, tc redeploy, or MCP set_env with redeploy: true) starts a new deployment from the current production image with the current variables, health-checks it and promotes it. No build, no scan, no migrations.

Limits​

LimitValue
Source bundle50 MB
Health-check timeout90 seconds
Primary services per projectOne web, api or static service; any number of cron and worker services
ProtocolHTTP
Was this page helpful?