Panelica MCP Server 0.5.1: connect Claude, Cursor, Codex and Gemini to your panel

[admin]

Panelica MCP Server 0.5.1 is here​

We just shipped the Panelica MCP (Model Context Protocol) server. If you've been wanting to point an AI assistant at your panel and have it actually do things instead of just talking about them, this is for you: create a domain, issue an SSL certificate, add a DNS record, spin up a database, check server health — all through plain-language requests to Claude, Cursor, Codex, or any other MCP-compatible client, wired straight into your panel's own External API.

What shipped​

• 404 tools across 51 categories — domains, DNS, SSL, email, databases, FTP, backups, Docker, security, WordPress, Git, and more. Every one of them is the same surface your panel UI already uses.
• Zero-drift catalogue. At startup, the server fetches your panel's own

  /v1/api-spec

  and builds the tool list from it. A new panel release means new tools automatically — nothing to hand-write, nothing to fall behind.
• A compact default. Rather than dumping all 404 tools on every client, the server registers a 34-tool core set plus 3 meta tools (37 total) by default — enough to stay under Cursor's roughly 40-tool active limit — with

  PANELICA_TOOLSETS=all

  available for clients that want the whole catalogue.
• Safety annotations on every tool. 181 read-only, 177 mutating, 46 destructive — clients like Cursor and Codex use these to auto-approve reads and ask for confirmation before anything destructive runs.
• Scoped API keys. 50 scopes across every area of the panel, nothing preselected when you create a key. Grant a read-only assistant

  *:read

  and it can look but not touch; grant

  domains:write

  and

  ssl:write

  for a deploy assistant and nothing more.
• The assistant knows the API "by heart." Server instructions injected on connect tell it how ids work (every id comes from a list call), what the response envelope looks like, what each error means and what to do about it, and its own key's scopes — so a read-only key gets announced as read-only up front instead of failing awkwardly mid-task.

Works with​

Claude Code, Claude Desktop, Cursor, and OpenAI's Codex CLI all have dedicated setup steps (linked below). Any other MCP-aware client that speaks stdio — Continue.dev, Cline, Zed, Google's Gemini CLI, or a custom integration — connects the same generic way: three environment variables and a command. That generic path matters more than it sounds: it means this server isn't tied to one vendor's assistant. Whichever MCP client you already use for coding, the same panel access travels with you — no separate plugin to maintain per client, no separate auth flow to relearn.

Install in 3 steps​

1. Generate a scoped API key. In your panel: Settings → API Keys → Generate API Key. Nothing is preselected — pick only the scopes this assistant actually needs, starting read-only if you're just exploring.
2. Point an MCP client at the server. Fastest path, no install required:
   

   npx -y panelica-mcp

   with three environment variables —

   PANELICA_BASE_URL

   ,

   PANELICA_API_KEY

   ,

   PANELICA_API_SECRET

   . Docker and build-from-source both work too if you'd rather not touch Node.js, or want to read the code first.
3. Ask it something. "List my domains" or "check the SSL status of example.com" is enough to prove the wiring end to end. Anything destructive is confirmed by the client before it runs, not fired blind.

Also open source​

MIT licensed. Full source, issue tracker, and the tool catalogue are on GitHub: https://github.com/Panelica/panelica-mcp. Package on npm: https://www.npmjs.com/package/panelica-mcp. A container image is published on every release to

ghcr.io/panelica/panelica-mcp

(tags

latest

,

0.5.1

). It's also listed in the official MCP Registry as

io.github.Panelica/panelica-mcp

, so any client that discovers servers from there picks it up automatically.

Prefer to run the AI agent inside the panel instead?​

Not everyone wants to install a CLI and manage an API key on their own machine. If you'd rather open a terminal, a live browser window, and a chat with Claude, Codex, Gemini or Cursor from inside the panel itself — no local install, plus a Telegram bridge for when you're not at a desk — that's OpsAI, a separate feature covered in its own guide, linked below.

Security, in one paragraph​

Every request is signed with HMAC-SHA256 over the method, path, timestamp and body using your local API secret — the secret itself never leaves the machine running the server, only the signature travels. Requests with a timestamp more than 5 minutes off the panel's clock are rejected outright, so replays don't work. The key you generate inherits its owner's role and the panel's normal RBAC, every call it makes lands in the panel's normal audit log traceable to that exact key, and keys carry a rate-limit tier (60 to unlimited requests/minute) so nothing can hammer the panel by accident. No telemetry, no caching, nothing phoned home.

Read next​

We split the documentation into two follow-up threads so this announcement stays short:

• Panelica MCP Server — complete guide (setup for every client, the full scope table, rate limits, 20+ example prompts by job, a real measured agent session, and troubleshooting): read it in API & Automation
• OpsAI: running Claude Code, Codex CLI, Gemini CLI and Cursor Agent inside the panel — if you'd rather run an AI agent from inside the panel itself, with its own live browser window and a Telegram bridge, instead of wiring up your own client: read it in General Discussion

There are also two shorter threads already up in this section: a 60-second Claude Code walkthrough, and a real transcript of Claude creating a site and issuing SSL end to end — both linked from the complete guide.

As of September 2026, version 0.5.1: 404 tools, 51 categories, 37 registered by default. Questions, bug reports, feature requests — post them here.