Documentation

MCP server

Ask about incidents and update them from your AI assistant, like Claude Code, Cursor or VS Code. Opsphie's Model Context Protocol server lets your assistant act as you: it sees what you can see, and what it changes shows under your name.

Connect your assistant

Sign in, open Personal settings from your account menu, and create an API token. Name it after the assistant and machine, like “Claude Code – laptop”, and choose its access: Read only, or Read and write to let it post updates and change incidents. Copy it: it's shown only once.

Add the server with your token. In Claude Code (the settings page shows this command with your token filled in):

claude mcp add --transport http opsphie https://api.opsphie.com/mcp \
  --header "Authorization: Bearer <your token>"

In Cursor, add it to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "opsphie": {
      "url": "https://api.opsphie.com/mcp",
      "headers": { "Authorization": "Bearer <your token>" }
    }
  }
}

In VS Code, add it to .vscode/mcp.json:

{
  "servers": {
    "opsphie": {
      "type": "http",
      "url": "https://api.opsphie.com/mcp",
      "headers": { "Authorization": "Bearer <your token>" }
    }
  }
}

Ask away: “What incidents are open?”, “Summarize the timeline of INC-42”, “Post an update on INC-42 that the rollback is out”.

Read

What you could see in the web app.

list_incidents: Incidents a page at a time, filtered by state, severity or status.
list_active_incidents: Incidents still being worked on, most severe first.
get_incident: One incident by number, with its details, people and full timeline.

Update

Each change goes on the timeline, and in Slack like one made with /inc.

post_update: Post a status update in the incident channel.
set_status: Change status, with a note on how it was resolved when resolving.
set_severity: Change severity, P0 to P3.
set_summary: Replace the summary on the control panel.
add_timeline_note: Add something that happened to the timeline, at the time it happened.

What your assistant can and can't do

Your assistant acts as you, through the tools above and nothing else. It sees your workspace's incidents, and private ones only if you're part of them (or an admin). It can't declare, delete or export incidents, or change settings. With a read-only token, it can only use the reading tools.

Its changes show on the timeline and in the incident channel under your name, marked via AI assistant, with the token's name on the timeline. Incidents are referred to by number: 42 for INC-42.

Opsphie doesn't call an AI model: your assistant does, with your own account, and it reads what the tools return.

Your tokens

Treat a token like a password. Opsphie keeps only a fingerprint of it, and tokens start with opsphie_ so secret scanners like GitHub's spot one committed by mistake.

Personal settings shows when each token was last used. Revoke one there and it stops working straight away. Tokens are also revoked when you're deactivated in Slack, or when your workspace uninstalls Opsphie.