Static and SPA hosting
Your site live with one command. No YAML.
You already built it. What is left is publishing, and that is where the workflow, the bucket, the DNS and the certificate show up. Hoster skips that part: one command, automatic HTTPS, and your SPA's /anything route does not return 404. With an MCP server, if you work with an agent. The dashboard is in Portuguese for now.
Before and after
Publishing should not be
A PROJECT.
Write the workflow, create the bucket, point the DNS, issue the certificate, find out why /dashboard returns 404 on reload. A whole afternoon for a site that was already working on your machine.
- No workflow file
- No bucket to configure
- No certificate to issue and renew
- No rewrite rule for your SPA
The whole CLI
We do not have a documentation site.
We have two commands, and they are the whole product.
1 · Sign in
One key, once
hoster login hk_sua_chave, and you are done. The key is saved in ~/.hoster-config.json, so you do not type it again. Nothing to install: bunx @ciromaciel/hoster-cli.
2 · Publish
hoster deploy --dir ./dist
That is the whole command. It zips, uploads and returns the URL of the live site with HTTPS. There is no step three.
3 · The flags
There are four. There is no fifth.
The whole CLI: --project sets the subdomain, --dir what to upload, --main the entry point and --token the access in CI/CD. All of them have defaults. No config file, no hoster.config.js, no hidden convention.
4 · CI/CD
The same command, in the workflow
No custom action to install, no plugin. It is the same line you run in the terminal, with --token coming from the repository secret.
For people who work with agents
Your agent publishes the site.
Without you pasting documentation first.
1 · Connect
One line, and the agent has the tools
It is a remote server: no package to install, no process to keep running. No key in the command: the client gets a 401, finds the login server on its own and opens the browser for you to approve.
$ claude mcp add --transport http \
hoster \
https://hoster.worker.myinfrastructure.click/mcp
→ 401 WWW-Authenticate
→ finds the login server
→ you approve in the browser
✔ Connected · 16 tools2 · Ask
In plain words, not in curl
The agent finds the projects, reads their state and acts. You do not paste documentation or explain the order of the steps.
You: "point site-cliente at the staging API" → hoster_read · hoster_get_project → hoster_write · hoster_set_variables ✔ no rebuild, no deploy
3 · The tools
Ten. One per intent, not per route.
The API has twenty endpoints; the server exposes ten tools. Publishing is one intent, not three calls, and the agent does not need to know the order.
list_projects get_project check_name check_domain set_variables set_cache set_domain purge_cache deploy_ticket deploy_status list_deployments rollback audit_project delete_project
4 · Publish
One sentence, and the site is live
The agent requests the ticket, runs the command and confirms. The files go straight from your disk to Hoster. They do not pass through the model, so publishing a site does not use up your context. The ticket is valid for one project, once, for fifteen minutes. Deleting a project requires explicit confirmation.
You: "publish the dist folder to site-novo"
→ hoster_write · hoster_deploy_ticket ✔ created
bunx @ciromaciel/hoster-cli push
ht_a1b2… --dir ./dist
→ hoster_read · hoster_deploy_status ✔ live
https://site-novo.myhoster.click5 · No key in any file
You approve in the browser
The connector stores no secret: the client finds Infrastructure Auth on its own from the 401, you authorize in the browser and the client keeps the token. Scope still applies. A read-only token does not even see the tools that delete.
$ claude mcp add --transport http \
hoster \
https://hoster.worker.myinfrastructure.click/mcp
→ authorize in the browser
✔ Connected · 16 tools
auth:read scope → only the 7 read toolsPricing
3, 9 or 99 projects
Two commands. No YAML.
If you get stuck on any step before you see your site live, that is on us, and we want to know which step it was.
FAQ
The questions we would rather answer up front
Is your case one we do not support?
Talk directly to engineering. If Hoster does not fit what you need, we tell you right away and point you to something that does.