Model Context Protocol
Your servers, one prompt away.
Connect Claude, ChatGPT, Cursor or any MCP client to Ploi. Deploy sites, read logs, create databases and restart queues — by simply asking.
Secured with OAuth · included in the Pro plan and up
How it works
Connected in under a minute
No API keys to copy around, no config files to hand-edit. Add the server, approve it once in your browser, and start asking.
Add the server
Point your MCP client at ploi.io/api/mcp — in Claude Code that is a single command.
Approve in your browser
A standard OAuth consent screen opens. You see exactly what access you are granting — and you can revoke it any time from your profile.
Just ask
Your assistant discovers your servers and sites on its own. Deploys, logs, databases, cronjobs — all a sentence away.
$ claude mcp add --transport http ploi https://ploi.io/api/mcp
Added HTTP MCP server ploi → run /mcp to authenticate
> Why did the last deployment of ploi.io fail?
# ChatGPT → Settings → Connectors → Create
Name ........... Ploi
MCP server URL .. https://ploi.io/api/mcp
Authentication .. OAuth
→ Sign in to Ploi and approve the connection
✓ Connected — enable Ploi under Tools in any chat
The toolbox
31 tools. Every daily job covered.
A curated set that mirrors the Ploi API — not a firehose. Read tools are marked read‑only for your assistant, mutations queue safely on your server, and destructive deletes are left out by design.
Servers
Sites
Deployments
Databases
Cronjobs
SSL certificates
Daemons & queues
Firewall
Everything runs through the same authorization as the Ploi API — your assistant can only see the servers your team can see.
Built-in guardrails
Powerful, not reckless
Giving an AI access to production infrastructure should never feel like a leap of faith. It doesn't.
- OAuth, not pasted keys
- Clients connect through a browser consent flow with PKCE. No long-lived secrets in config files.
- Granular scopes
- Prefer an API token? Every tool honours the same read/create scopes as the REST API.
- Team scoped
- Your assistant only sees the servers and sites your teams have access to. Nothing else exists.
- No deletes, ever
- Destructive delete operations simply don't exist in the tool set. Removing things stays a human decision.
- Fully auditable
- Deployments triggered by an assistant show up in your history tagged “via MCP”, next to every other action.
- Revoke in one click
- Every connected application is listed in your profile. One click disconnects it — instantly.
Just say it
Things you can ask
“Why did the last deployment of ploi.io fail?”
Reads the deployment log and points at the failing step
“Add a cronjob that runs the Laravel scheduler every minute”
create-cronjob, installed on the server within seconds
“The queue seems stuck since the last deploy — fix it”
Finds the site's workers and restarts them
“Request an SSL certificate for shop.example.com”
request-certificate via Let's Encrypt, status checked for you
“Only allow 10.0.4.20 to reach MySQL on the staging server”
create-firewall-rule for port 3306 with a source IP
“Compare the .env of staging and production”
get-site-env on both sites, differences summarised