MCP Server

Back in August, Tom (our CMO) released our prototype MCP server. It was open source, ran locally, and let you talk to incident.io from AI tools like Claude and Cursor.

Many customers used it (a lot, in fact) and so it was clear the demand was there, but the feedback around it was consistent:

"We love the idea, but we can't adopt something that runs locally, and isn't officially supported."

Pretty fair. So, we built the real thing - an officially supported, remote MCP server with all the power of the original, and then some.

It's been in Private Beta for the past few weeks and the response from customers has been excellent. Today, as we continue to iterate and improve it, we're opening up a Public Beta to all paying customers.

What's new

No local setup required, no repo to clone. Just connect your AI tools of choice and authenticate with user based OAuth (to inherit user permissions) or use an API key (if you'd like to connect from internal tools). You'll be up and running in under a minute.

It works seamlessly with Claude Desktop, Claude Code, Cursor and any MCP-compatible client you may care to use and you can allowlist additional domains if you’re making use of an MCP proxy like Cloudflare, too.

What can you do with it?

You get full access to incidents, alerts, on-call schedules, catalog, and workflows. We've also provided tools that can speak with our Agent (which you could already use via @incident in Slack, or via our Web Dashboard), to tap into our intelligence layer directly over MCP.

Ask questions like:

• "What happened overnight?"
• "Show me all incidents affecting the payment service in the last 30 days"
• "Draft my on-call handover notes to post in #handovers in Slack"
• "Who's on call for the platform team right now?"
• When did we last have an incident with Postgres?
• "Declare a new high severity incident called 'Database performance degradation’"

If you have early access to AI SRE (currently in Private Beta) you can also "Talk to Production" and run incident investigations, working directly with our Agent to debug live incidents with all the context in your account alongside telemetry, logs and errors. For example:

• "Investigate why we're seeing elevated error rates on the checkout service"
• "What's the current load on our production database and how does that compare to normal?"
• "Have any of my recent deployments affected error rates in production?"

Get started in 2 minutes

1. Head to Settings > MCP to enable the MCP server for your organisation.
2. Head to the docs to configure your AI tool(s) of choice
3. Authenticate via OAuth when prompted

That's it. The full power of the incident.io platform is now available for you to use over MCP!

What's next

If you've been waiting for a supported, hosted MCP integration - this is it. Go try it out, and let us know what you think!

This is a public beta which means we're still actively iterating on it based on your feedback. Some things may change as we develop new use cases, improve our tooling coverage and more. Please give us feedback directly through the MCP, via email, or in your shared Slack support channels.

We're also in the process of releasing our native app for macOS, which comes packaged with MCP support out of the box. If you'd be keen to test that out, let us know here: https://app.incident.io/~/settings/desktop-app

Managing API keys programmatically

That’s not all we shipped this week - organizations can now manage their API keys programmatically via the new API Keys V1 public API. This means you can now do things like:

• Rotate API keys according to a schedule
• Creating / managing a large number of API keys (ie. for each team in your organization)

Bonus: You can now rotate an API key's token, meaning you can continue using the same key (same name, permissions) but with a fresh token! Just click on the overflow menu for any API key to leverage this!

What else we’ve shipped