Introducing the Panelica MCP Server: 404 Tools to Manage Your Hosting in Plain English

The Panelica MCP server is an official Model Context Protocol server that exposes 404 tools covering the entire hosting panel — accounts, domains, databases, DNS, email, SSL, backups, Docker, application hosting, Git and more — so any MCP-compatible AI assistant can manage your server in plain English instead of clicking through menus. It installs with a single command and talks to your panel through the same signed External API that already powers Panelica's own dashboard.

What it is

MCP — the Model Context Protocol — is an open standard that lets AI assistants call external tools in a structured, discoverable way instead of guessing at APIs from documentation. The Panelica MCP server is our implementation of that standard for hosting: it takes the panel's External API, described in a live OpenAPI specification, and turns every relevant endpoint into a named tool an AI client can call directly. It is published on npm as panelica-mcp, as a container at ghcr.io/panelica/panelica-mcp:latest, and listed in the official MCP Registry as io.github.Panelica/panelica-mcp. The source is public at github.com/Panelica/panelica-mcp.

Installing it

No local install step is required. Any MCP client that can run an npx command can start the server on demand:

npx -y panelica-mcp

Or run it as a container if you prefer to manage it as a long-lived process:

docker run --rm -i \
  -e PANELICA_BASE_URL=https://your-panel:8443/api/external \
  -e PANELICA_API_KEY=pk_... \
  -e PANELICA_API_SECRET=sk_... \
  ghcr.io/panelica/panelica-mcp:latest

Either way, the server needs three environment variables: the base URL of your panel's External API, an API key and a matching secret. We walk through generating those and wiring up specific clients in How to Manage a Web Server with Claude Code and How to Connect Codex CLI to Your Hosting Panel via MCP.

What 404 tools actually cover

The number is not padding. It reflects the panel's real External API surface of 406 endpoints, grouped into families:

• Accounts & RBAC — create, suspend, update and inspect hosting accounts and their permissions.
• Domains & DNS — provision domains and subdomains, manage zone records, check propagation.
• Databases — create and manage MySQL and PostgreSQL databases and users.
• Email — mailboxes, forwarders, autoresponders, DKIM/SPF/DMARC configuration.
• FTP / files — accounts, quotas and direct file operations.
• SSL — issue, renew and inspect certificates.
• Backups — trigger, list and restore backups.
• Cloudflare — zone and DNS sync for connected accounts.
• Docker & application hosting — containers, images, compose projects, and the Laravel, Node.js and Python runtimes.
• Git & deploy — repository management and deploy workflows, 43 endpoints on their own.
• Logs — 26 endpoints for reading service and application logs.
• Security — firewall rules, IP blocking, and the built-in Security Advisor.
• Server & services — service status and control across the 20 services Panelica manages.

Every one of these families is covered because the MCP catalogue is generated from the panel's own live /v1/api-spec, not hand-maintained. We explain exactly how that generation works, and why the catalogue can only grow, in How We Keep 404 AI Tools in Sync with a Live API.

Safety is built into the tool, not bolted onto the client

Two mechanisms keep an AI assistant from doing something you did not intend. First, the API key you generate has 50 scoped permissions — across accounts, domains, databases, DNS, email, FTP, SSL, backups, files, Cloudflare, Docker, apps, Git, logs, security and server management — and none of them are preselected; you search and pick exactly what the key can do. Second, every tool carries a safety annotation of read-only, mutating or destructive: 181 tools only read state, 177 change something, and 46 are destructive operations like deleting a database or removing a domain. MCP clients use these annotations to prompt for confirmation before anything destructive runs. We cover this in depth in Is It Safe to Give an AI Access to Your Server?

What using it feels like

Once connected, you stop opening the panel for routine work. Asking an AI client "create a new PostgreSQL database for the staging app and give me the credentials" or "show me the last 100 lines of the Node.js app's error log" resolves in one exchange, using the exact same authenticated API calls the panel's own dashboard makes — not a shell wrapper, not a scrape of the UI. For a full walkthrough of a real multi-step task, see Create a WordPress Site by Talking to an AI.

Get started