Overview
The Buildkite AI connector gives Rootly AI current CI/CD evidence during an investigation. Rootly AI can inspect pipelines and builds, trace a failed build to its jobs and logs, check artifacts and annotations, and correlate Test Engine failures with an incident. The connection uses Buildkite’s hosted, read-only Model Context Protocol (MCP) server. You authorize Rootly from Buildkite through OAuth; you don’t create or paste an API token. Rootly always connects to Buildkite’s dedicated read-only endpoint and can’t trigger, retry, cancel, or modify pipelines and builds.Before You Start
You’ll need:- A Buildkite user account with access to each organization and pipeline Rootly AI should investigate.
- Permission to authorize OAuth applications in Buildkite.
- Permission in Rootly to manage AI connectors.
Connect Buildkite
Open the Buildkite AI connector
Authorize Rootly in Buildkite
Confirm the connection
What Rootly AI Can Read
The reviewed Buildkite tool set covers these investigation surfaces:read_, plus tools that require no scope. Rootly also pins the reviewed tool names, so an upstream tool added later doesn’t become available until Rootly reviews it.
Investigation Examples
Try questions that start broad and then narrow to the failing job or test:- “Which Buildkite build first failed before this incident, and what changed from the last passing build?”
- “Summarize the failed jobs in the latest build. Show only the log lines that support the diagnosis.”
- “Did the build promise a failure before the job finished? Check the final job result and any later logs.”
- “Are the same tests failing across recent Test Engine runs, or is this failure new?”
- “Did an agent or queue problem delay this build?”
- “Find artifacts and annotations from the failed build that could explain the deployment regression.”
failing while a job is still running when the job has declared a promised failure. Rootly AI can begin investigating that signal, then wait for the terminal result and check for additional logs, artifacts, or test results.
Access and Data Handling
- No personal API token: You don’t create or paste a Buildkite API access token. Rootly stores the OAuth access and refresh credentials encrypted so it can use and refresh the connection.
- Read-only endpoint: Rootly fixes the connection to
https://mcp.buildkite.com/mcp/readonly. - Permission-scoped: Rootly AI sees only data available to the Buildkite user who authorized the connection.
- Reviewed tools: Rootly limits the connector to its pinned, read-only tool set.
- Investigation-time queries: Rootly queries Buildkite when the investigation needs CI/CD evidence.
- Redacted session history: Raw Buildkite tool results are available to the active investigation but aren’t persisted in the AI session transcript.
- Quality traces: As with other Rootly AI tool calls, connector results can appear in retained large language model (LLM) traces used for quality monitoring. See Data Privacy for Rootly AI for the retention boundary.
Troubleshooting
Buildkite denies the OAuth request
Buildkite denies the OAuth request
The connector is connected but a pipeline is missing
The connector is connected but a pipeline is missing
Rootly AI can't find artifacts, schedules, agents, or Test Engine results
Rootly AI can't find artifacts, schedules, agents, or Test Engine results
A dynamic step upload can't be found
A dynamic step upload can't be found
Buildkite queries fail after the connector worked previously
Buildkite queries fail after the connector worked previously
Frequently Asked Questions
Can Rootly AI trigger, retry, or cancel a Buildkite build?
Can Rootly AI trigger, retry, or cancel a Buildkite build?
Can I connect with a Buildkite API token?
Can I connect with a Buildkite API token?
Can Rootly AI read multiple Buildkite organizations?
Can Rootly AI read multiple Buildkite organizations?
How do I limit which Buildkite data Rootly AI sees?
How do I limit which Buildkite data Rootly AI sees?