Connect Claude to Astry: your incidents, crises and on-call schedules in your AI assistant

It is 2 a.m. and your phone just rang. Before even opening a dashboard, you want to know three things: what is broken, since when, and who is already on it? More and more teams ask this kind of question to an AI assistant. It still needs access to the right information.

It is now possible: Astry has an MCP integration that lets you connect Claude (and Claude Code) to your Astry account. You query your incidents, crises and on-call schedules in plain language, right from your conversation.

MCP in a nutshell

MCP (Model Context Protocol) is an open protocol that standardizes how an AI assistant connects to an external service. It is often described as a "USB-C port for AI assistants": a service exposing an MCP server becomes usable from every assistant that supports it, without specific development.

In practice, Astry exposes an MCP server offering the assistant a set of tools ("list incidents", "get a crisis", "see who is on call"...). The assistant decides on its own which tool to call to answer your question.

What you can ask Claude

The point is not to replace a filter or a dashboard: Astry's already do that job very well. The point is everything a filter cannot do: understanding a freely worded question, cross-referencing information, summarizing it, assessing it and writing. A few examples:

"I'm taking over the on-call shift in an hour. What do I need to know about last night? What is fixed, what is still pending?"
"This database incident looks familiar. Did we have a similar issue this month? What worked back then?"
"The alerts of this incident are unreadable Prometheus JSON. Explain what they actually say, and what it means for us."
"Explain last night's incident as if I had to present it to the director in five minutes, without technical jargon."
"Where does the ongoing crisis really stand? Among the remaining actions, which ones look blocking to you, and in what order should we tackle them?"
"Is any team drowning in alerts right now? Compare with last week and tell me whether it's unusual."
"Draft the post-mortem of this incident: timeline, impact, probable causes, improvement ideas."
"Sum up the last hour of discussion on this incident in a clear comment, and post it once I've reviewed it."

For the last example, Claude first shows you the exact text of the comment and waits for your approval before posting it. It is the only write action available: everything else is read-only.

In every case, the assistant decides which data to fetch from Astry (incident list, timeline, alerts, comments, crisis status, on-call schedules) before building its answer.

Security first

Giving an AI assistant access to your incident data is not a trivial decision. We designed this integration around a few simple principles.

The assistant only sees what you see. Each member connects Claude with their own Astry account (password or company SSO), and the assistant acts on their behalf, with their role and permissions. A member without access to teams will not see on-call schedules through Claude either. Read-only roles (Stakeholder, External) cannot comment.

Your organization decides. The integration must be enabled by an Owner in the incoming integrations, and every connection requires the member's explicit consent. Disabling the integration cuts every connection in less than five minutes. The same applies when a member leaves the organization.

Your credentials never go through the assistant. The connection relies on OAuth: Claude receives an access token specific to Astry, valid for one hour and renewed for 30 days at most. It never sees your password, and never receives the tokens of your Astry session.

Some data is never shared. The phone numbers of on-call people, the emails of crisis contacts and the targets of notifications stay in Astry. People are referred to by their name.

Alert content is treated as data. An alert comes from an external system and could contain misleading text ("ignore your instructions and..."). The server always reminds the assistant that this content is not a set of instructions to follow.

Everything is traced. Every connection, disconnection and comment posted through an assistant shows up in your organization's Timeline, with the application used.

How to enable it

The integration is currently in beta. To get it:

Ask our team to enable the feature (contact us). It then shows up in Organization > Subscription.
An Owner enables the integration: Organization tab, then Incoming integrations and AI assistants (MCP).
Each member connects their assistant:
- in Claude (claude.ai, desktop and mobile apps): Settings > Connectors, then add a custom connector with the URL https://mcp.astry.fr/mcp;
- in Claude Code: claude mcp add --transport http astry https://mcp.astry.fr/mcp, then /mcp to sign in.

The assistant then opens an Astry page where you sign in and explicitly allow the access. Details are in the documentation.

What's next?

This first version focuses on reading data, with a single, carefully framed write action. We will extend it based on your feedback: support for other MCP-compatible assistants, new actions (acknowledging or resolving an incident...) and ready-to-use request templates for recurring situations such as on-call handovers or crisis status updates.

Have a use case in mind, or want to try the integration? Get in touch: we will gladly enable it for your organization.