New: Connect AI assistants via MCP

Integrations

Connecting AI Assistants (MCP)

Connecting AI Assistants (MCP)

Sentinel ships a built-in server that speaks the Model Context Protocol (MCP), the open standard for connecting AI assistants to external tools and data. Because it's standard MCP, it works with any MCP-compatible client: Claude, ChatGPT, and other agents and IDEs. Once connected, you can ask things like "what needs my attention this morning?", "is anything down right now?", "which sites are slowest this week?", or "check if example.com is up from every region" in plain language, and the assistant reads the answer straight from your account.

Team-scoped, with read and write access by plan

An assistant only ever sees the team you're currently working in, and any plan with API access can use the read tools. Write tools (create, update, delete, pause monitors and manage incidents) require full API access (Pro or Business) and are refused otherwise; your assistant asks you to approve each write, and deletes are flagged destructive. See the MCP Tools reference for everything it exposes.

Connection endpoints

Sentinel offers two ways to authenticate, depending on the client:

  • OAuth (for hosted assistants like the Claude and ChatGPT apps), one-click sign-in, no token to copy:
https://sentinel.rootstuff.io/mcp/sentinel-oauth
  • Bearer token (for developer tools like Claude Code, Claude Desktop, or your own agent), send an API token as an Authorization header:
https://sentinel.rootstuff.io/mcp/sentinel

Claude

Claude app (web/desktop), on a paid Claude plan:

  1. Open Settings → Connectors → Add custom connector
  2. Name it "Sentinel" and paste https://sentinel.rootstuff.io/mcp/sentinel-oauth
  3. Click Add and approve the consent screen

Claude Code:

claude mcp add --transport http sentinel https://sentinel.rootstuff.io/mcp/sentinel --header "Authorization: Bearer YOUR_TOKEN"

Claude Desktop, add to your MCP config file:

{
  "mcpServers": {
    "sentinel": {
      "type": "http",
      "url": "https://sentinel.rootstuff.io/mcp/sentinel",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

ChatGPT

ChatGPT supports custom MCP connectors on paid plans (Plus, Pro, Business, Enterprise). Add a connector in ChatGPT's Settings → Connectors, choose to add a custom/remote connector, and enter the OAuth URL https://sentinel.rootstuff.io/mcp/sentinel-oauth, then authorize when prompted.

Note

ChatGPT's connector UI evolves and may require enabling developer mode to access an MCP server's full tool set. If the menu differs from the above, follow OpenAI's current connector instructions and use the same OAuth URL.

Other MCP clients

Any MCP-compatible client (other agents, IDE extensions, or your own code) can connect over HTTP. Point it at https://sentinel.rootstuff.io/mcp/sentinel with an Authorization: Bearer YOUR_TOKEN header, or at https://sentinel.rootstuff.io/mcp/sentinel-oauth if it supports OAuth. Then list and call the tools documented in the MCP Tools reference.

Autonomous agents and scoped tokens

Agents that run unattended (scheduled jobs, CI pipelines, background assistants) should get their own API token rather than reusing yours, so it can be revoked independently. When you create a token on the API settings page you choose its permissions, and Sentinel enforces them: a token with only read can use every read tool but is refused by every write tool, even on a Pro or Business plan. Grant create, update, or delete only if the agent genuinely needs them; an agent that triages incidents needs update (acknowledge, resolve, pause) but not delete. A refused call returns an error naming the missing permission, so the agent can report exactly what to change instead of failing silently.

Good to know

  • The token authenticates the assistant to Sentinel only, it has nothing to do with your AI subscription.
  • Hosted-app connectors generally require a paid plan on the AI provider's side; the bearer-token method works with any MCP client.
  • The read tools are safe to "always allow" so routine questions like "is anything down?" skip approval prompts; write tools (Pro/Business) will still ask you to confirm each change.
  • Usage is tracked alongside your other API activity, so you always know what's being queried.
WebhooksMCP Tools