MCP Integration -- Query Monitoring from Your Chat

Your Monitoring Becomes Part of Your Chat

Instead of opening the dashboard, you ask your AI assistant:

“Which of my checks are currently red?” “What was the uptime of shop.my-company.com last month?” “Show me the last five incidents, sorted by severity.”

The assistant fetches the answer live from your live24h workspace – no copy-pasting, no tab-switching, no numbers pulled from memory.

What is MCP?

Model Context Protocol – an open standard (Anthropic, November 2024) that lets AI assistants access external data sources in a controlled way. live24h provides an MCP server that you run locally and connect to your assistant. The data stays on your machine and inside your workspace; the AI provider only sees what you send it in the chat.

What Kind of Questions Can You Ask?

The live24h MCP server exposes five read-only tools:

Tool	Answers questions like…
`get_checks`	“Which checks do I have at all?”, “Which ones are paused?”
`get_check_status`	“Is `www.my-company.com` up right now?”, “When was the last run?”
`get_incidents`	“What are my most recent incidents?”, “Is an incident ongoing?”
`get_status_pages`	“Which public status pages have I published?”
`get_uptime_stats`	“What was my uptime in April?”, “How many minutes of downtime last week?”

Read-only, no writes – the assistant cannot create, change or delete checks. That stays deliberately reserved for you and the web interface.

Setup – Claude Desktop

1. Create an API Key in live24h

Dashboard -> Settings -> API Keys -> New Key. Set the scope to your workspace. The key is shown only once – copy it immediately and store it securely (password manager).

2. Install the MCP Server

At the moment we ship the server as an open-source component. You clone the repository and build it once:

git clone https://github.com/live24h/mcp-server.git
cd mcp-server
dotnet publish -c Release -o ./publish

Coming soon: A ready-made dotnet tool package and a Docker image, so this step shrinks down to a single line.

3. Adjust the Claude Desktop Configuration

Open the configuration file:

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

Add an entry for live24h under mcpServers:

{
  "mcpServers": {
    "live24h": {
      "command": "C:\\pfad\\zu\\publish\\Brixtaq.MonitoringService.McpServer.exe",
      "env": {
        "LIVE24H_API_KEY": "sk_live_xxxxxxxxxxxxxxxxxxxxxx",
        "LIVE24H_WORKSPACE_ID": "ws_01HP..."
      }
    }
  }
}

On macOS/Linux, use the absolute Unix path to the binary without the double backslashes.

4. Restart Claude Desktop

After the restart, the sidebar shows an entry called live24h-monitoring with a green dot and the label “5 Tools”. Ready to go.

5. Give It a Try

Ask: "Which of my live24h checks are currently red?"

Claude calls get_checks in the background, filters for error status and replies in full sentences.

Setup – Cursor

Cursor supports MCP via the file ~/.cursor/mcp.json (Windows: %USERPROFILE%\.cursor\mcp.json). The JSON format is identical to Claude Desktop – just copy the block above. See the Cursor documentation under Settings -> MCP for details.

Example Dialogue

You:    Which of my checks are offline right now?

Claude: Your check "shop.my-company.com" has been in an error state for
        7 minutes (timeout after 30 seconds). All other 12 checks are
        green. Shall I open the corresponding incident?

You:    Yes, please. And what was the uptime of that shop in March?

Claude: March 2026: 99.84 % uptime for shop.my-company.com. That is
        71 minutes of downtime across 3 incidents. The longest outage
        lasted 38 minutes on March 14.

Security

Your machine stays the only connection point. The MCP server runs as a local process on your computer – not in the cloud. It only talks to api.live24h.eu. The AI provider only sees the tool results that Claude/Cursor themselves request.
Workspace scope. An API key is tied to a single workspace. For multiple workspaces, create multiple MCP server entries with different keys (e.g. live24h-prod, live24h-staging).
Whitelist enforcement. Only five read-only tools are enabled. An architecture test in the backend prevents additional tools from being activated by accident.
Rate limit. 60 calls per minute globally, 10 per minute for statistics-heavy tools.

Availability

MCP access is available from the Business plan upwards. On smaller plans you can build and run the MCP server locally, but you will receive 403 Forbidden on the first tool call.

Troubleshooting

Problem	Cause	Solution
Red dot in Claude	Wrong path to the binary	Enter the absolute path; on Windows use double backslashes
`401 Unauthorized` in the log	API key expired or wrong workspace	Generate a new key in the dashboard
`403 Forbidden`	Plan too small	Upgrade to Business or request access
No tools visible	Server started but tool list empty	Rebuild the binary from the current version
“Workspace ID not provided”	No default workspace ID set	Add `LIVE24H_WORKSPACE_ID` to the config

What’s Next?

One-line install via dotnet tool install live24h-mcp (in progress)
Docker image for setups without a .NET SDK on the host
HTTP/SSE transport in addition to the local stdio transport – for cloud-hosted AI assistants
Additional tools such as get_notification_rules, get_audit_log – added step by step on purpose, to keep the attack surface small