Ask your AI assistant how your sites are doing
A-Line.cloud has an MCP server. MCP clients such as Claude, Codex, Cursor and VS Code connect to your account, look up your apps, deploys, build logs, domains, backups and bill, and answer in plain words. You sign in from the client and decide which organisations it may see. It is included at no extra charge.
Fix a build, merge it, watch it go live
You: the dentist-site build failed after my last push, can you fix it?
a-line - list_builds (MCP)(app: "dentist-site", limit: 3)
b_214 failed in the build step, 07:58Running
a-line - get_build_log (MCP)(build: "b_214")
Read 48 linesRunning
[ERR_MODULE_NOT_FOUND] Cannot find package '@astrojs/sitemap' imported from /app/astro.config.mjs
Update(package.json)
Added @astrojs/sitemap to dependenciesRunning
Bash(pnpm build)
Built 24 pagesRunning
Bash(gh pr create --title "Add the sitemap package")
Opened pull request #87Running
astro.config.mjs imports @astrojs/sitemap, but it was never added to package.json, so the build stopped. It builds here now, and pull request #87 is open. Shall I merge it once the checks pass?
You: yes, merge it when it's green
Bash(gh pr checks 87 --watch)
All checks passedWaiting for checks
a-line - list_actions_jobs (MCP)()
test ran on a-line-4vcpu in Germany, 1m 12sRunning
Bash(gh pr merge 87 --squash)
Merged into mainRunning
a-line - list_builds (MCP)(app: "dentist-site", limit: 1)
b_215 from main, builtBuilding
a-line - list_deploys (MCP)(app: "dentist-site", limit: 1)
Release 42 liveRunning
Merged. The checks ran on your A-Line runners and passed in 1m 12s. dentist-site deploys on every push to main, so build b_215 started on the merge, and release 42 is live.
Check on your client sites
You: Anything I should know about my client sites this morning?
a-line get_overview12 apps: 11 running, 1 build failedRunninga-line list_buildsFailed at 07:58, nothing releasedRunning11 of your 12 apps are running as usual. The dentist-site build from 07:58 failed, so the site still serves yesterday's version: visitors see no difference.
You: Is the bakery's new domain live yet?
a-line list_domains1 domain, waiting for DNSRunningNot yet. The domain doesn't point to A-Line.cloud, so its certificate is waiting. Add the DNS record the console shows for it at the registrar, and the certificate follows on its own.
You: And is the webshop's database backed up?
a-line list_backupsDaily at 03:00, 14 keptRunningYes. shop-db is backed up every night at 03:00 and the last 14 backups are kept, in the EU. Last night's finished at 03:04.
Roll back two apps, then find the bug
You: Checkout throws errors since this morning's deploy. Roll the shop back, then find what broke.
list_deploys app: "shop-web", limit: 210:42 live, 08:15 before itRunninglist_deploys app: "shop-api", limit: 210:40 live, 08:12 before itRunningredeploy_previous app: "shop-web"Back on 08:15Runningredeploy_previous app: "shop-api"Back on 08:12RunningGrep "total" src/3 matches in checkout.tsRunningEdit src/lib/checkout.tsSends amount instead of totalRunningBoth apps are back on this morning's earlier releases, so checkout works again. The cause: the 10:40 API release renamed total to amount, and the shop still sent total. I've fixed checkout.ts.
You: Looks right, push it.
Bash git pushPushed to mainRunningget_app_status app: "shop-web"Running the new releaseDeployingget_app_status app: "shop-api"Running the new releaseRunningPushed. Both apps built from main and run the new releases, with the fix and the renamed field together.
Find what drives the CI bill, then fix it
You: our Actions spend jumped this month, what's going on?
Called a-line.get_actions_usage(period: "30d")
e2e: 68% of the minutes, on a-line-16vcpuRunning
Called a-line.list_actions_jobs()
e2e ran 214 times, on pushes to every branchRunning
Read .github/workflows/e2e.yml
Runs on every push, to any branchRunning
Most of it is e2e. It runs on a 16 vCPU runner for every push to every branch, so a branch with an open pull request is tested twice for each push. Running it on pull requests and on main only would drop about half of its runs. Shall I change that?
You: yes, open a PR for it
Edited .github/workflows/e2e.yml (+4 -2)
Runs on pull requests and pushes to mainRunning
Ran gh pr create --title "Run e2e once per change"
Opened pull request #31Running
Ran gh pr checks 31 --watch
All checks passedWaiting for checks
Pull request #31 is open and green. Once it's merged, e2e runs once per change instead of on every push, on the same runner.
Ask what you would look up in the console
31 tools, live today. Each reads your account the moment the assistant asks, within the organisations you granted and with your role in each.
Ask it things like
- Is anything down this morning?
- Why did last night's dentist‑site build fail?
- Is the bakery's new domain live yet?
- When was the webshop's database last backed up?
- Who changed the shop's settings yesterday?
- Which workflow eats most of our Actions minutes?
All 31 tools, by area
- Apps and status
get_overviewfind_appget_app_statusget_platform_status- Builds and deploys
list_buildsget_build_loglist_deploysget_deploy- Domains and certificates
list_domains- Config, storage and backups
get_app_configlist_volumeslist_backupslist_dynamic_secrets- Actions runners
list_actions_jobsget_actions_usage- Billing, for owners and admins
get_billing_summaryget_billing_insightslist_transactionslist_invoices- Activity
list_activity
Clusters, networks, load balancers, usage, organisation settings and the config-as-code status are there too.
You decide what it can see
You connect by signing in to A-Line.cloud from the client. API keys aren't accepted, so there is none to paste into a config file or leak in a repository.
- It acts as you
- With your role in each organisation: billing needs the owner or admin role, as it does in the console.
- Only the organisations you pick
- Connect one client's organisation and leave the rest out. An organisation can turn MCP off, and its two-factor rule applies here too.
- Secrets stay secret
- Secret values never come back, only their names, and build logs are redacted first.
- Revoke it any time
- Under Connected apps in your account settings. An organisation's admins can end any member's connection.
Next: it does the work too
The next step lets the assistant deploy, roll back, scale and change settings when you ask, within limits you set and the platform enforces. Until then it reads, and you make the change in the console.
It will deploy, cancel a deploy, roll back, restart, scale, stop and start apps, set environment variables and secrets, add domains and take a backup on demand. New apps, deletes, restores and node pool changes it only drafts, for a person to approve.
With limits built in
- You pick which kinds of change each connection may make, per organisation and environment. Allowing them takes a fresh sign-in.
- Hourly limits per app, such as six deploys an hour, which an organisation can only lower.
- Every change shows up in the activity feed.
- A retried request returns its first result instead of running twice.
Connect in a minute
It works with MCP clients that connect to remote servers and sign in over OAuth. The address is the same everywhere:
https://mcp.a-line.cloud
- Add the address to your client, with the snippet for it.
- Your browser opens A-Line.cloud. Sign in and pick the organisations it may see.
- Ask it how your sites are doing.
Claude Code
claude mcp add --transport http \
a-line https://mcp.a-line.cloudYour browser opens to sign in the first time Claude Code uses it.
Codex
codex mcp add a-line --url https://mcp.a-line.cloud
codex mcp login a-lineThe second command opens your browser to sign in.
Cursor
{
"mcpServers": {
"a-line": { "url": "https://mcp.a-line.cloud" }
}
}Your browser opens to sign in the first time Cursor connects.
VS Code
{
"servers": {
"a-line": { "type": "http", "url": "https://mcp.a-line.cloud" }
}
}Start the server from the file and sign in when VS Code asks.
Claude
Name a-line
URL https://mcp.a-line.cloudIn Claude Desktop or Claude on the web. Claude opens A-Line.cloud to sign in when you connect.
Questions people ask
Something else on your mind?
Write to us and Arthur replies, within 24 hours on working days.
Which assistants work with it?
MCP clients that connect to remote servers and sign in over OAuth, such as Claude Code, Claude Desktop, Codex, Cursor and VS Code.
What does it cost?
Nothing extra: MCP is included with A-Line.cloud. Once the assistant can make changes, those cost what they would from the console, such as the build minutes of a deploy.
Does the AI model see my data?
It sees the answers to the tools it calls, and those go to the model your client uses. Which provider that is, and where it runs, depends on the client: check its terms. A-Line.cloud runs no AI model itself.
Can it change or delete things?
Not yet: today it reads. Deploying, scaling, restarting and changing settings from the assistant are coming, only where you allow them. It will never delete anything directly: deletes, restores and new apps it only drafts, for a person to approve in the console.
We run sites for many clients. Does one connection see them all?
Only if you grant them all. You pick the organisations per connection, so you can connect one client's organisation and leave the rest out.
How do I disconnect it?
Under Connected apps in your account settings, where you end the connection.
How do I get in?
A-Line.cloud is invite-only while we take on new customers one at a time. Here is what happens.We reply within 24 hours on working days.
Already hosted somewhere else? During early access, we do the move for you.
Today
You tell us who you are
A few lines about you and what you want to run.
Within 24 hours on working days
We reply
With questions, or with your invite.
After your invite
You go live
Sign in, and we help you with your first deployment.