AskYourStack MCP server
Let Claude, ChatGPT, Cursor or Claude Code manage your own Linux servers, with approvals for anything risky.
AskYourStack is a hosted, remote MCP server. Install a small agent on your server with one line, add your private AskYourStack MCP address to your AI client, and ask in plain words: set up a raw VPS, install Magento or WordPress, fix a 500 error, take backups, clean up malware, harden the server or move a site to a new one. For SEO it reads Google Search Console, Google Analytics 4, Core Web Vitals, on-page audits and your structured data next to what Googlebot really fetches in your server logs, and fixes the causes on the server.
- Website: https://askyourstack.com
- Docs: https://askyourstack.com/docs
- Security model: https://askyourstack.com/security
- Prompt library: https://askyourstack.com/prompts
This repository holds the listing and documentation for the AskYourStack MCP server. The service itself is hosted at askyourstack.com; there is nothing to build or run locally.
How it works
- Connect your server. Sign up free at askyourstack.com, name your server and paste the one-line install as the Linux user who owns your sites (the agent then manages only that user's sites and files), or as root when it should run the whole server. The agent dials out to askyourstack.com over HTTPS: it opens no port, and no SSH key or password is stored anywhere.
- Add AskYourStack to your AI client with your private MCP address from the dashboard (see below).
- Ask. Your AI looks at the server first, explains its plan, and runs the work through AskYourStack's tools. Anything destructive waits for your approval.
Connect your AI client
Two ways in:
- Sign in (OAuth): clients that support sign-in for remote MCP servers connect to
https://askyourstack.com/mcp. You sign in to AskYourStack, see which app is asking and allow it. Each app gets its own access, which you can end under Security in the dashboard. - Private address:
https://askyourstack.com/mcp/YOUR-ADDRESS, from the dashboard, for clients without sign-in. Keep it private, and make a new one in the dashboard if it leaks. The examples below use it.
The tool and playbook catalogue is public at https://askyourstack.com/mcp/catalog.
Claude (claude.ai and Claude Desktop): Settings, Connectors, Add custom connector. Name it AskYourStack, paste the address and leave the OAuth fields empty. Guide
Claude Code:
claude mcp add --transport http sudowhizzy https://askyourstack.com/mcp/YOUR-ADDRESS
Cursor (mcp.json):
{
"mcpServers": {
"sudowhizzy": { "url": "https://askyourstack.com/mcp/YOUR-ADDRESS" }
}
}
ChatGPT: Settings, Connectors, turn on developer mode, then add a custom connector with your address and no authentication. Guide
Any other client that supports remote MCP servers over streamable HTTP works the same way.
Tools
| Tool | What it does | Plan |
|---|---|---|
overview | Where everything stands in one call: plan, servers and their sites, open problems and warnings from the monitoring, approvals waiting, connected Google properties, saved notes and recent work | Free |
propose_plan | Ask approval once for a job with several risky steps: the user sees every risky step and approves them together; each then runs once, exactly as listed | Free |
wait_for_approval | After giving the user an approval, plan or change-window link, wait in the same turn until they decide, then carry on without them returning to the chat | Starter |
save_note, forget_note | Short notes about a server or the account that the next chat sees in overview (never passwords: those are refused) | Free |
log_work | A plain summary of a finished task, for the owner's dashboard and the next chat | Free |
list_servers | The servers on the account: online, safety mode, paused | Free |
server_facts | Distro, CPU, memory, disk, package manager, services, ports, Magento and WordPress installs | Free |
get_playbook | Step-by-step guides for the AI (see Playbooks) | Free |
health_report | Disk and inodes, memory, load, failed services, certificates, security updates, reboot needed, backups (age and rhythm), out-of-memory kills, PHP out of workers, database connection limit; on Pro and Agency also software with a known vulnerability | Free |
list_sites | The sites on the server: Magento 2, OpenMage (Magento 1), WordPress and WooCommerce, PrestaShop, Shopware 6, Drupal, Joomla, OpenCart and Laravel, found from the web server's own configuration (Plesk, CloudPanel, RunCloud, DirectAdmin and plain layouts) and in Docker containers: kind, root, domains, access logs, owner, version, database (passwords never shown), its console | Free |
logs | A log by name (web-error, php, magento, wordpress, database, system, mail, auth, or any file) for a time window, with repeated errors counted once, a sample of each and the latest lines | Free |
Search Console history: all 16 months Google keeps, on every plan (Free reads one property). Google Analytics 4: Pro 10 properties and 90 days, Agency 25 and all history. Core Web Vitals: 100 checks a day on Starter, 300 on Pro, 1,000 on Agency. Agency can connect up to 3 Google accounts (clients' own).
Playbooks
Served as MCP prompts and through get_playbook: server-setup, magento, wordpress, hardening, backups, site-checkup, malware-cleanup, migrate, bot-attack, seo-technical, publish-content, performance, updates, site-down, add-site, staging, restore, magento-upgrade, email, woocommerce, prestashop, shopware, drupal, laravel, other-shops (Joomla, OpenCart, OpenMage), seo-dashboard, whizzycommerce, ghost. The AI fetches the right one on its own for tasks like a fresh Magento install or a hacked site.
Safety
- Every call is sorted by risk. Reads run at once. Changes run after
/etcis saved. Destructive actions (deleting your data, dropping databases, SSH, sudo and firewall changes, reboots) wait for your approval. Anything touching the agent's own key is blocked. - Approvals the AI cannot fake. You approve signed in on askyourstack.com; an approval runs once, only with the exact arguments you saw, within 30 minutes. Text planted in a log or web page cannot approve anything. The approval page says in plain words what the action does, what could go wrong and how to undo it, and before an approved delete of database tables or whole folders a copy is saved first when the command names them plainly.
- Modes and kill switch. Each server is read-only, normal or full trust. Pause a server or the whole account in one click; disconnecting revokes the agent.
- Undo.
/etcsnapshots before changes, previous versions of every overwritten file, snapshots and rollback. - Secrets stay on the server. Database credentials are read from the site's own config and never reach the AI. Your MCP address is stored only as a hash.
- Signed agent updates and a full audit log of every call.
More at https://askyourstack.com/security
Supported servers
Linux with systemd (Debian, Ubuntu, Rocky Linux, AlmaLinux, RHEL and similar), x86_64 or ARM64, from any provider. A fresh VPS or a server already running a shop or site.
Pricing
Free: one server, read-only, 50 tool calls a day, and Google Search Console for one property with the opportunities report. Paid plans add changes, approvals and snapshots and the SEO tools with Google Search Console (Starter, 1 server, 3 properties), the malware scan, site checks every 15 minutes, Google Analytics 4, on-page and technical SEO tools and longer history (Pro, 5 servers, 10 properties; Agency, 25 servers, 25 properties, up to 3 Google accounts). Current prices: https://askyourstack.com/#pricing. Every tool with what it does and its plan: https://askyourstack.com/docs/tools. Your AI client is billed separately by its own provider.
Support
info@askyourstack.com · Terms · Privacy
AskYourStack is operated by Whizzy Digital Solutions Lda, Portugal. The agent installed on your server is open source (Apache-2.0): https://github.com/shopwhizzy/askyourstack-agent, with a script that checks each release against its source. The hosted service is proprietary; this README may be quoted in MCP directories and listings.