Changelog
ChangelogStatus PageAPI ExplorerAPI ReferenceCLI ReferenceBlog
Back to All
Added

Connect your own MCP servers to Monte Carlo's agents

about 2 hours ago

You can now connect your own MCP servers to Monte Carlo, like Notion, Linear, GitHub or Datadog.

Bayes, Monte Carlo's AI teammate, can then call their tools to bring in context Monte Carlo doesn't collect, such as incidents, tickets and runbooks.

Bayes uses your MCP servers in two places:

  • Chat: Ask Bayes a question and it can look things up in your connected tools in the same conversation. For example, it can find the Linear ticket for a pipeline change or the runbook for a failing job, so you don't have to switch tools and paste it in.

  • Troubleshooting alerts: When Bayes troubleshoots an alert, it can pull in related incidents, tickets, code changes and logs as evidence. More evidence means a better root-cause analysis.

Set up:

  • Admins add a server in Settings → MCP servers with a name and URL, and Monte Carlo detects the auth settings from the URL.

Supported auth:

  • No auth

  • API key, with extra headers when needed

  • OAuth client credentials, for shared machine-to-machine access

  • OAuth 2.1 per user: each user connects their own account from the MCP servers menu in chat, and Bayes only sees what that user can see

Bayes only calls tools the server marks as read-only.

Note: Troubleshooting alerts currently works only with servers that use shared credentials (no auth, API key or OAuth client credentials). Servers that use per-user OAuth are available in chat only for now.

Learn more here: Custom MCP servers | Technical overview