AGENTIC-SETUP

title: Agentic Setup — Wire UpMonitor to Claude Code, Cursor, Claude Desktop description: Step-by-step MCP server install for every major AI agent. One-time setup. Every Active Incident from then on becomes a copy-paste handoff. keywords: MCP server uptime monitoring, Claude Code MCP uptime setup, Cursor uptime monitor integration, Claude Desktop MCP, ChatGPT MCP infrastructure monitoring ogTitle: Agentic Setup — Wire UpMonitor to Your AI Agent ogDescription: 4 install paths. 30 seconds each. Then every incident is a one-paste fix.

Agentic Setup — Wire UpMonitor to Your AI Agent

Connect the UpMonitor MCP server to your AI agent once, and every Active Incident panel becomes a one-paste fix. This guide covers the four supported agent surfaces — Claude Code, Cursor, Claude Desktop, and ChatGPT — with the exact config snippet for each. Total install time per agent: under 30 seconds.

The UpMonitor MCP server is published as @upmonitor/mcp-server on the npm registry. It runs locally on your machine via npx, holds your UpMonitor API key in environment scope, and exposes tools the agent calls on demand. Configuration is identical across surfaces — only the wrapper format differs per host application.

Prerequisites: get an API key

You need exactly one credential before installing into any tool: your UpMonitor API key.

  1. Sign in at upmonitor.io/login.
  2. Open Developer Settings → API Keys.
  3. Click Generate Key, copy the value (format: up_...), and paste into the config snippets below.

Keys are scoped to the user account that issued them. The MCP server inherits that scope — it cannot mint new keys, escalate privileges, or read other users' monitors.

Claude Code is the fastest install because the claude mcp add subcommand handles the config write for you.

claude mcp add upmonitor \
  -e UPMONITOR_API_KEY=up_... \
  -- npx -y @upmonitor/mcp-server

Verify the connection by running claude mcp list and confirming upmonitor appears with status connected. Then ask Claude Code: "Use the upmonitor MCP server to list my monitors." The first call may take 3-5 seconds while npx fetches the package; subsequent calls are sub-second.

Install path 2: Cursor

Cursor stores MCP servers in Settings → MCP → Add Server. Paste this JSON block into the editor that opens:

{
  "upmonitor": {
    "command": "npx",
    "args": ["-y", "@upmonitor/mcp-server"],
    "env": {
      "UPMONITOR_API_KEY": "up_..."
    }
  }
}

Reload the Cursor window after saving. The server appears under the MCP Servers section of the chat sidebar with a green dot when active.

Install path 3: Claude Desktop

Claude Desktop reads its MCP config from claude_desktop_config.json. The file location depends on your OS:

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

Add this block under mcpServers:

{
  "mcpServers": {
    "upmonitor": {
      "command": "npx",
      "args": ["-y", "@upmonitor/mcp-server"],
      "env": {
        "UPMONITOR_API_KEY": "up_..."
      }
    }
  }
}

Restart Claude Desktop entirely (quit + relaunch — reload is not enough). The UpMonitor hammer-icon tool button appears in the chat composer once the connection is live.

Install path 4: ChatGPT (custom GPT or Desktop with MCP)

ChatGPT's MCP support landed via the Desktop app's Connectors pane. Open Settings → Connectors → Add MCP Server, then paste:

{
  "name": "upmonitor",
  "command": "npx",
  "args": ["-y", "@upmonitor/mcp-server"],
  "env": {
    "UPMONITOR_API_KEY": "up_..."
  }
}

For browser-based ChatGPT users without MCP support, the Active Incident handoff still works — paste the prompt directly into the chat. The agent loses the structured MCP context but retains the human-readable root cause and remediation steps the prompt carries.

What the agent can do after install

Once installed, the agent gains access to the UpMonitor tool surface. The current toolkit covers the highest-value verbs; the MCP integration guide lists the full schema.

  • run_audit(url, checkers?) — runs SSL, DNS, HTTP, performance, and security-header probes against any URL on demand.
  • get_usage() — reports the agent's current API quota and rate-limit window.
  • get_active_incident(monitor_id) — coming next release — returns the structured incident object the Active Incident panel renders, so the agent can reason over root cause + remediation without parsing the prompt.

The AI agent then drives the diagnosis loop end-to-end: it can re-run individual checkers to verify a fix, fetch usage telemetry to advise on quota, and (in the next release) accept incident IDs directly to skip the prompt-paste step entirely.

Verifying the install

After installing into any tool, run this one-line smoke test in the agent chat:

*"Use the upmonitor MCP server to run an audit on `upmonitor.io` and tell me the SSL certificate's days-to-expiry."*

A correct install returns a structured answer within 5-10 seconds, including the issuer, protocol version, and exact daysRemaining integer. If the agent reports the server is not available, recheck the API key scope and restart the host application.

Common install issues

The two install failures users hit most often have explicit error signatures and 30-second fixes.

MCP server not found after install — the host application was not restarted (Claude Desktop) or the window was not reloaded (Cursor). MCP server registration runs at host startup, not on config save.

401 Unauthorized on the first tool call — the UPMONITOR_API_KEY env var either contains a stale value or was pasted with surrounding whitespace. Open Developer Settings, regenerate the key, and re-run the install command (which overwrites the env scope).

Security model

The MCP server holds your API key in process-local environment scope. It is not written to disk by the server itself and is not transmitted anywhere except upmonitor.io on outbound API calls. To rotate, revoke the old key in Developer Settings and re-run the install command for each agent.

The MCP server's outbound traffic targets https://upmonitor.io/api/v1/* exclusively — there is no telemetry, analytics beacon, or third-party endpoint. Audit the source at github.com/upmonitor/mcp-server if you want to verify before running npx.