Your server. Your AI of choice. Full DevOps superpowers.
A self-hosted agent that exposes 32 DevOps tools to any MCP-compatible client — and runs its own built-in chat against any model you bring. No SaaS. No lock-in. No monthly bill.
subscription or API key — Anthropic, OpenAI, xAI, or Google · Installs in 60 secondsThree steps to an AI engineer with server access
Install the agent
Upload the PHP files, open install.php, follow the wizard. Works on cPanel, Plesk, VPS, Docker — any PHP 7.4+ host. No database required.
Connect your AI client
Add the agent as a connector (OAuth, one click) or paste an MCP token into Claude Desktop, Claude Code, Cursor, Zed… 32 new tools appear in your sessions instantly.
Just talk to your AI
“Check git status on prod.” “Tail the Laravel log.” “Back up, then deploy.” Your AI calls the right tool — using the subscription or API key you already pay for.
Built to outlive the hype cycle
Auto-detects active models
Queries each provider’s live model list with your key. If a model is retired, the agent switches to the closest active one automatically and logs it — no manual updates, ever.
Security that ships by default
OAuth 2.1 + PKCE for MCP, bcrypt logins, TOTP 2FA, AES-256 encryption of keys and tokens at rest, role-based access, full audit log.
No phone-home
License verifies offline with a cryptographic seal. If our site vanished tomorrow, your install keeps working. Your data never leaves your server.
Shared-host friendly
Flat-file storage, no database, no Node, no Docker required. Written to survive real-world cPanel hosts — malware scanners, disabled functions and all.
All 32 DevOps tools, one MCP server
Every tool your AI engineer needs, callable from a single conversation — in whatever client you prefer.
Shell & Files 5 tools
Git & Database 4 tools
System & Services 5 tools
Network & Logs 4 tools
Ops & Safety 8 tools
Memory & Context 6 tools
Shell commands are allowlist-gated · file writes auto-backup · every call lands in the audit log · roles decide who can call what.
Nothing else does this
Control panels aren’t agents. IDE agents don’t know your server. Generic MCP servers don’t do ops.
| Product | Self-hosted | AI agent | MCP server | BYO AI provider | One-time license |
|---|---|---|---|---|---|
| Server Agent (this) | ✓ | ✓ | ✓ 32 tools | ✓ 4 providers | ✓ from $29 |
| Laravel Forge / Ploi / RunCloud | SaaS only | ✕ | ✕ | N/A | $12–30/mo |
| cPanel / Plesk | ✓ | ✕ | ✕ | N/A | License+ |
| Continue / Cline / Aider | IDE-only | ✓ | ✕ | API only | Free / API |
| Generic MCP servers | varies | ✕ | ✓ narrow | ✓ | Free |
Simple pricing. Own it forever.
One-time purchase. No subscriptions, no phone-home, no usage limits. Install on your server, keep it.
- All 32 MCP tools
- Web admin panel + built-in chat
- Per-user MCP tokens + OAuth connect
- Unlimited team users
- Persistent knowledge base
- Lifetime updates
- Multi-server hub mode
- White-label branding
- Everything in Starter
- Hub mode: control 5 servers from one MCP client
- Cross-server tool calls (deploy main, restart workers)
- Auto-issued seat tokens to remotes
- Unified audit log across servers
- Priority email support
- White-label branding
- Everything in Pro
- Unlimited servers (no seat cap)
- White-label branding (name, tagline, logo, color)
- Bulk MCP token issuance
- Priority support + setup help
30-day refund · Secure checkout via Gumroad · License emailed instantly
Also on CodeCanyon: Starter (Regular License) · Agency (Extended License)
Questions
What if I don't have an AI subscription or API credits?
You'll need some way to run an AI model — either a paid chat subscription or pay-as-you-go API access. The agent supports four providers: Anthropic, OpenAI, xAI, and Google. Pick whichever you already use, or sign up for the cheapest plan that fits your usage. There's no free path that avoids both — but you only pay one provider, not us.
Does this burn my API credits?
It depends on how you use it. MCP mode (recommended): connect a desktop client that uses your chat subscription, and inference runs through that subscription — no API calls, no credits. Built-in chat mode: the agent calls your provider's API directly using your API key (and that does consume credits). Most users go MCP-first; built-in chat is the fallback for environments without a desktop client.
What happens when a provider retires a model?
Nothing you need to do. The agent queries each provider's live model list with your key (cached 24h), so the Settings panel always offers current models. If the model you configured is retired mid-conversation, the agent detects the error, switches to the closest active replacement, retries, and logs the swap in the audit trail.
Which AI providers and clients work?
Providers for built-in chat: Anthropic, OpenAI, xAI (Grok), Google (Gemini). MCP clients tested: Claude Desktop, Claude Code CLI, Cursor, Zed. Any MCP-compatible client should work — we use the standard JSON-RPC over HTTP transport, with OAuth or Bearer tokens.
Is MCP allowed by these AI vendors' terms?
Yes — MCP (Model Context Protocol) is an open standard for tool servers. Connecting an MCP server to a desktop client is the officially sanctioned extension mechanism. We don't proxy auth tokens or scrape session cookies (those would violate ToS). We expose tools through the standard MCP interface, which is exactly what it was designed for.
What kind of server do I need?
Any PHP 7.4+ host with file-write permissions. Tested on shared cPanel, Plesk, DirectAdmin, Docker, AWS EC2, DigitalOcean, Hetzner, Linode, and generic VPS. The installer auto-detects your host and warns about limitations before you install.
What's the difference between this and Laravel Forge?
Forge provisions servers and handles deploys — it's a SaaS control panel. This is an AI agent you talk to through your AI client. It reads your code, tails your logs, runs queries, debugs issues. Also: Forge is $12/mo/server forever. This is $29 once.
Can my team share one install?
Yes — all paid tiers include unlimited team users (admin/developer/viewer roles). Each member generates their own MCP token; revoke individual tokens when someone leaves.
What happens if I don't renew?
Nothing. It's a one-time license with lifetime updates. No subscription, no expiry. We don't phone home. If our site disappeared tomorrow, your install keeps working.
Can I white-label it for clients?
White-label (custom product name, tagline, logo, color scheme) is exclusive to the Agency tier — every surface (login screen, dashboard, agent header, MCP server name) picks up your brand.
How secure is it?
OAuth 2.1 with PKCE for MCP connections, bcrypt password hashing, AES-256 encryption of API keys and server tokens at rest, TOTP 2FA, IP allowlist, per-user role-based permissions, CSRF tokens, rate-limited logins, and auto-rename of the installer after setup. Every action lands in the audit log.
What if I need help?
All paid tiers ship with the full setup guide and HTML docs. Pro and Agency tiers include priority email support; Agency also includes setup assistance.