Skip to main content
Extensions are currently in Private Beta. Get in touch by email or our shared Slack channel if you’d like to try them out.
Investigations build their own understanding of how your organization works. We learn from your incidents, your code, and your telemetry, and we keep that picture current so nobody on your team has to maintain it. Extensions are for the parts you’d rather hold yourself: the internal systems we have no way to reach, and the times you want to say how a particular kind of incident should be handled rather than leave us to work it out. You give investigations your own tools and your own instructions, and both get used at the right moment in the same investigation as everything else.

​
Two ways to extend

Plugins are content investigations read. A plugin is a directory of skills in one of your repositories. A skill is a SKILL.md file that explains how to do something in your environment: how to triage a failing checkout, what to check when the job queue backs up, how your services depend on each other. Investigations load a skill when it matches the work in front of them, then follow it. Connectors are systems investigations call. A connector is a server you’ve connected that exposes tools an investigation can use, so it can query something we have no native integration for. Connectors speak MCP, an open standard for letting AI systems use your tools. The rule for choosing between them: content you want investigations to follow is a plugin, a system you want investigations to query is a connector. Most teams end up with both, because the two work together. A skill is the procedure, a connector is what carries it out.
Plugins use the Claude Code plugin format, so skills you write for us also work in your own tooling. You aren’t writing something that only we can read.

​
How investigations use this

In investigations. Skills are loaded during an investigation when they fit what it’s trying to understand, or at a fixed point you choose if you want one to run every time. A skill doesn’t run in isolation: the agent following it can query your telemetry, search your code and docs, and call your connectors, with the incident and everything the investigation has found so far in front of it. Connector tools are called the way any other source is queried, and what comes back becomes evidence in a finding. In the agent. Ask @incident something during an incident and it can reach for the same skills and the same connectors, so a question in the channel can draw on your systems too.

​
What you can build

​
Reach a system we don’t support yet

You run something central to your stack that we don’t integrate with natively yet. Connect it as a connector, then write a skill that says how to debug with it. A team running Temporal might connect their Temporal MCP server, then write a skill explaining which workflows matter, what a stuck workflow looks like, and which failures are worth escalating. Investigations go from having no view of Temporal to reading it the way that team does.

​
Make investigations follow your runbook

This one isn’t about access. Your team already knows how a particular kind of incident should be worked, and you’d rather we followed that than found our own route each time. Write your triage procedure as a skill and investigations follow it. If checkout incidents always start with the payment provider’s status and the queue depth, say so, and that’s where the investigation starts too.

​
Setting up

Everything is set up from Extensions in your dashboard.

Plugins

Sync skills from one of your repositories.

Connectors

Connect an MCP server and choose the tools we can call.

​
FAQs

The native integration, always. Our telemetry data sources are tuned for telemetry work in a way a generic tool call can’t match. Use a connector for systems we don’t integrate with natively.
No. Extensions work with the agent as well, so you can connect your tools and skills and use them through @incident without running investigations.
They work together, and most teams want both. Connecting your docs lets us search everything your team has written, wherever it lives. A plugin bundles a chosen set of files and hands them straight to the agent, so what you want followed is already in front of it rather than something it has to go looking for.
Not on their own. A connector tool that makes changes can only be called when someone is talking to the agent, and only after an administrator has explicitly allowed that tool. See Connectors.
The investigation carries on and does the best it can with what it has. You don’t have to notice this yourself: we review how each skill gets used and flag when one reached for something that wasn’t there. See seeing how a skill is used.

Connect your data

The sources investigations draw on, and how to set each one up.

How investigations work

How evidence becomes a finding.