MCP server

Vantaj ships a hosted Model Context Protocol server, so AI assistants can work with your monitoring directly - ask Claude "which of my monitors are down?", have Cursor create a monitor for the endpoint you just built, or let an agent check incident history while debugging.

https://api.vantaj.co/mcp

Authentication uses your regular API key - MCP access requires a paid plan, same as the REST API.

Connecting a client

Claude (claude.ai / Claude Desktop)

Add a custom connector under Settings → Connectors:

URL: https://api.vantaj.co/mcp
Header: Authorization: Bearer vtj_live_your_key

Claude Code

claude mcp add --transport http vantaj https://api.vantaj.co/mcp \
  --header "Authorization: Bearer vtj_live_your_key"

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "vantaj": {
      "url": "https://api.vantaj.co/mcp",
      "headers": { "Authorization": "Bearer vtj_live_your_key" }
    }
  }
}

Smithery

The server is also listed on Smithery, where you can connect through the Smithery gateway or try the tools in their playground - supply your Vantaj API key as the apiKey parameter.

Clients that only support local (stdio) servers

Bridge with mcp-remote:

{
  "mcpServers": {
    "vantaj": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.vantaj.co/mcp", "--header", "Authorization: Bearer vtj_live_your_key"]
    }
  }
}

Available tools

Tool	What it does
`list_projects`	List projects - call first, most tools need a `project_id`
`list_monitors`	Monitors in a project with status, response time, uptime
`get_monitor`	One monitor with full settings
`create_monitor`	Create a monitor (checks start immediately) - supports all types incl. inverted checks
`update_monitor`	Change name, URL, interval, inverted flag, and other settings
`pause_monitor` / `resume_monitor`	Stop and restart checks for a monitor
`create_maintenance_window`	Suppress alerts for chosen monitors during a planned window - ideal before deploys
`list_heartbeats`	Heartbeat monitors with last-ping status
`list_incidents`	Open and resolved incidents with duration
`get_incident`	One incident by ID
`list_domains`	Tracked domains with expiry countdowns
`list_status_pages`	Status pages with slugs and visibility

What's deliberately not possible

MCP clients are AI models, and models can be manipulated - so the MCP surface is reads plus additive writes only. There are no delete tools, and alert policies, integrations, team, and billing settings aren't exposed. Anything destructive stays in the dashboard and the REST API, where a human is in the loop.

Scopes, limits, and behavior

Tool calls run through the same pipeline as the REST API: your key's scopes apply (monitors:read, monitors:write, …), plan limits are enforced on creation, and calls count toward your key's rate limits.
The server is stateless Streamable HTTP - no sessions, no OAuth dance, just your API key as a bearer header.
Responses are the same JSON shapes as the REST API, so anything an assistant learns from the API docs applies.