Connect StillOnline MCP in popular AI IDEs

You shipped a health URL and external probes are running. The annoying part is opening the dashboard every time you want to know if production is green—or asking the agent to add a check after a deploy without leaving the editor.

Model Context Protocol (MCP) exposes StillOnline as tools your agent can call: list projects, read last_status on checks, pull the public status JSON. Keys stay in local MCP config, not in chat.

This guide covers Cursor, Windsurf, Google Antigravity, OpenAI Codex CLI, Claude Code, Claude Desktop, and any other MCP client. Package: stillonline-mcp on npm (Pro/Ultimate).

Quick answer

stillonline-mcp (Pro/Ultimate) lets AI agents in Cursor, Windsurf, or Claude call StillOnline: get_public_status, list_checks, create HTTP monitors from the IDE. Run npx -y stillonline-mcp setup, add sk_live_… from API settings—never in chat. MCP reads the same API as REST guide; it does not replace owner alerts.

What you can ask the agent

Once stillonline appears in your MCP server list, treat StillOnline like any other tool the agent can call—no need to paste API responses into chat by hand. The table maps common questions to tools; the prompts below are copy-paste starters.

After setup, typical liveness / uptime workflows:

Full tool list: /en/docs/mcp. Under the hood these call REST API v1 at https://api.stillonline.tech/v1.

Example prompts:

• “Call StillOnline get_public_status for slug my-saas and summarize if anything is down.”
• “List my StillOnline projects, then list_checks on the production project — flag any last_status not OPERATIONAL.”
• “After this deploy, create an HTTP check on /api/health expecting 200.”

Do not paste sk_live_… into the prompt — only in MCP env.

Prerequisites

MCP is optional glue on top of checks you already run in the dashboard. You need a paid StillOnline plan, a key created once in the browser, and Node on the machine where the IDE runs.

1. Pro ($9/mo) or Ultimate — Free has no API key and no MCP. Pricing.
2. Create sk_live_… at API settings (shown once; store in a password manager).
3. Node.js on your machine — MCP runs npx -y stillonline-mcp over stdio.

Optional env: STILLONLINE_API_BASE if you proxy the API (default is production).

Fastest path: setup wizard

In Terminal or PowerShell (not in AI chat):

npx -y stillonline-mcp setup

Russian CLI: add --ru. PowerShell paste issues with hidden key input → --visible-key.

The wizard:

1. Confirms the npm package
2. Lets you pick your IDE (arrow keys)
3. Writes config or prints a CLI command
4. Validates the API key against the live API

Target a client directly:

npx -y stillonline-mcp setup --client cursor
npx -y stillonline-mcp setup --client windsurf --yes

Then reload MCP in the editor (Cursor: Settings → MCP → Refresh).

Per-IDE configuration

Universal JSON block (manual paste):

{
  "mcpServers": {
    "stillonline": {
      "command": "npx",
      "args": ["-y", "stillonline-mcp"],
      "env": {
        "STILLONLINE_API_KEY": "sk_live_..."
      }
    }
  }
}

Cursor

• Config: ~/.cursor/mcp.json (Windows: %USERPROFILE%\.cursor\mcp.json)
• Wizard: auto-writes the file
• Reload: Settings → MCP → enable stillonline → Refresh

Windsurf (Cascade)

• Config: ~/.codeium/windsurf/mcp_config.json
• Wizard: auto-writes
• Reload: Windsurf MCP panel → restart server

Google Antigravity

• Config: Antigravity MCP JSON (wizard resolves path on your OS)
• Wizard: auto-writes
• Same mcpServers.stillonline shape as Cursor

OpenAI Codex CLI

Codex uses TOML, not JSON. Recommended:

codex mcp add stillonline --env STILLONLINE_API_KEY=sk_live_... -- npx -y stillonline-mcp

Or append to ~/.codex/config.toml:

[mcp_servers.stillonline]
command = "npx"
args = ["-y", "stillonline-mcp"]
enabled = true

[mcp_servers.stillonline.env]
STILLONLINE_API_KEY = "sk_live_..."

Verify: codex mcp list — stillonline should appear.

Claude Code

claude mcp add stillonline --scope user --env STILLONLINE_API_KEY=sk_live_... -- npx -y stillonline-mcp

• --scope user → all projects on this machine
• --scope project → team .mcp.json without committing secrets

Restart Claude Code; test with list_projects.

Claude Desktop

Same JSON as Cursor, different file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop after editing.

Other MCP clients

Any client that supports stdio + npx works. Paste the universal JSON into the agent’s MCP config, reload, call list_projects to verify. Docs: /en/docs/mcp.

Verify liveness from the agent

1. Public status (no key on this call): ask for get_public_status with your status page slug (from the dashboard URL or /s/{slug}).
2. Owner detail: list_projects → pick project_id → list_checks → read last_status and last_probed_at.

If a check was paused (enabled: false), last_status may still show OPERATIONAL — that is the last probe before pause, not live health.

Pair MCP with a real health route first: Health check quickstart. For scripts and CI without an IDE, use the REST API guide.

MCP reads and configures checks; it does not replace owner alerts. When a probe fails, StillOnline can still notify you via email, Telegram, or Slack (Telegram setup · Slack setup) — one channel on Free, all three on Pro / Ultimate in account settings.

Limits and security

MCP does not add a second probe network—it calls the same API your scripts use. Keep keys out of chat logs and respect rate limits when agents loop in a tight retry.

• MCP configures and reads external HTTP probes — it does not replace them.
• Rate limits: 120 GET / 30 write per minute per API key; 429 includes retry_after.
• Rotate keys at API settings if one leaks into a shared transcript.
• npx install never asks for your key — only the setup wizard on your machine.

Related guides

• Health check URL quickstart.
• REST API: read uptime status.
• Telegram owner alerts · Slack alerts.
• OpenClaw agent monitoring.

FAQ

Does StillOnline MCP work on the Free plan?

No. StillOnline API keys and stillonline-mcp require Pro ($9/mo) or Ultimate—see Pricing. On Free, use the web UI and public status JSON if the page is public; upgrade for MCP and private API reads.

Which StillOnline MCP tool answers “is my API up?”?

Use get_public_status for the customer-facing slug (no key on that tool). For owner detail—probe time and OPERATIONAL / DOWN—use list_projects then list_checks. Same data as REST API; full tool list in /en/docs/mcp.

Is stillonline-mcp the same in Cursor, Windsurf, and Claude?

Yes. One npm package (stillonline-mcp) and the same tools; only the MCP config file path differs per IDE. Run npx -y stillonline-mcp setup --client cursor (or windsurf, etc.) after creating a key in API settings.

Can StillOnline MCP read a private status page?

No for anonymous public JSON. get_public_status works for public slugs only. Private visibility returns 404 on the public endpoint—use authenticated list_checks with your API key or sign in to the StillOnline app.

Is stillonline-mcp the same as OpenClaw MCP?

No. OpenClaw runs its own gateway MCP server. stillonline-mcp talks only to StillOnline’s uptime API. For OpenClaw gateway health, publish a public /health URL and monitor it—OpenClaw guide.

What if the StillOnline MCP setup wizard rejects my API key?

From package 0.3.3+, an empty or invalid sk_live_… shows an error and prompts again. Create a fresh key in API settings; use --visible-key in PowerShell if paste fails. MCP does not replace Telegram or Slack owner alerts.