AI ASSISTANTS (MCP)

Your incident platform, inside the assistant you already use.

OpsKnight is a Model Context Protocol server. Connect Claude, ChatGPT, Cursor, VS Code or any MCP client with per-person sign-in, scoped permissions and human-reviewed changes.

WHAT IT SOLVES

Each person connects their own assistant.

People sign in with their OpsKnight account, including single sign-on, and approve what the assistant may do. Assistants act with that person's current permissions, never more, and role or team changes apply on the next request.

Understand the capability
01Per-person OAuth sign-in
02Read incidents, on-call and services
03Reviewed, audited changes
PRODUCT WORKFLOWAI assistants (MCP)

OpsKnight is a Model Context Protocol server. Connect Claude, ChatGPT, Cursor, VS Code or any MCP client with per-person sign-in, scoped permissions and human-reviewed changes.

HOW IT WORKS

A concrete operational path, not a feature list.

MCP SERVERAssistants act as the person who connected them, and changes wait for review.
01Assistant
02OAuth sign-in
03Person's permissions
04Review
05Audit log
Per-person sign-inChanges off by defaultAudited

EXPLORE A WORKFLOW

  1. 01Add server URL
  2. 02Sign in to OpsKnight
  3. 03Approve permissions
  4. 04Connected

Each person adds the MCP server URL to their assistant, signs in with their own OpsKnight account and approves what the assistant may do.

OPERATIONAL DEPTH

AI assistants (MCP), beyond the happy path.

The details below are the parts teams need when evaluating how the capability behaves during real response, failure, and handoff.

01

Ask about incidents, on-call and services.

Catch up on what paged you, investigate an incident with its context and similar past incidents, find who is on call and who is next, check a service's health and recent changes, or draft a postmortem from the timeline.

Read the operational guide
02

Changes are off until an administrator allows them.

Routine incident changes such as acknowledge, assign, note and resolve then run within the person's own permissions. Configuration changes and runbook runs are prepared, frozen and wait for a human reviewer before they apply.

Read the operational guide
03

Know what reaches the AI vendor.

Reviewed fields only, secrets always removed, and a request log that never stores arguments or results. Consent, revocations, approvals, committed changes and sensitive reads land in the audit log.

Read the operational guide
04

Works with remote and local clients.

Remote clients connect over OAuth with automatic registration. CLIs and headless servers can use device sign-in, and clients that cannot reach a remote server use the opsknight-mcp stdio bridge.

Read the operational guide
05

Build your own agent.

Connect your own code with the MCP SDK: page through results, retry safely with idempotency keys and drive the approval loop.

Read the operational guide

KNOW BEFORE PRODUCTION

Validate the boundary, not just the happy path.

Use a test service and representative provider configuration before treating AI assistants (MCP) as production incident infrastructure.

MCP access is off until an administrator turns it on, and changes stay off until they are allowed. Each client's tested status and limits are listed in its setup guide.

  1. 01Turn on AI assistant access with read-only defaults and connect one assistant from a test account.
  2. 02Confirm a Responder's assistant cannot read or do more than that Responder can.
  3. 03Allow reviewed changes, then prepare and approve one configuration change and check the audit entry.
DOCUMENTATIONSetup, authorization, limits, and troubleshooting.
Explore AI assistants (MCP) documentation

Your incidents should belong to you.

Run OpsKnight on infrastructure you control.

v2.2.0 · AGPL-3.0-only · Self-hosted · 28 inbound integrations