Buildkite MCP Server
Connect AI agents to your Buildkite CI/CD pipelines, builds, jobs, and tests over MCP.
Updated
What is Buildkite MCP?
The Buildkite MCP server is a Model Context Protocol server that exposes Buildkite CI/CD data — pipelines, builds, jobs, logs, artifacts, and Test Engine results — so AI agents and editors can read and act on your continuous integration workflows.
Built and maintained by Buildkite, it interacts with the Buildkite REST API and ships in two flavors. The hosted remote server at https://mcp.buildkite.com/mcp uses OAuth, issuing short-lived tokens so you never have to manage API keys by hand; a read-only variant (/mcp/readonly) and an API-token pass-through endpoint (/direct) are also available. Alternatively, you can self-host the open-source server locally (the Buildkite docs recommend running it in a container) using a Buildkite API access token.
With the server connected, an agent can list pipelines, trigger and cancel builds, retry or unblock jobs, search build logs, inspect artifacts, and pull Test Engine run data — turning Buildkite into a queryable, actionable tool for AI-driven DevOps and CI/CD automation.
Tools the Buildkite MCP server exposes
| Tool | What it does |
|---|---|
| current_user | Get details about the user account owning the API token. |
| user_token_organization | Fetch the Buildkite organization associated with the token. |
| list_pipelines | List all pipelines in an organization. |
| get_pipeline | Get detailed pipeline configuration and statistics. |
| create_build | Trigger a new build on a pipeline. |
| cancel_build | Cancel a running build. |
| list_builds | List all builds for a pipeline. |
| get_build | Retrieve detailed build information including its jobs. |
| retry_job | Retry a failed or timed-out job. |
| unblock_job | Unblock a blocked job to allow execution to continue. |
| search_logs | Search a build's log entries using regex patterns. |
| list_artifacts_for_build | List a build's artifacts across all jobs. |
| get_failed_executions | Retrieve failed Test Engine execution data with details. |
Connect the Buildkite MCP server
Claude Code (remote, OAuth)
- 1
Add the remote server
Run: claude mcp add --transport http buildkite https://mcp.buildkite.com/mcp
- 2
Authorize via OAuth
On first use, Claude Code opens a browser to authorize Buildkite; approve the requested scopes. Tokens are short-lived and refresh automatically.
- 3
Verify the connection
Run /mcp in Claude Code and confirm the buildkite server is connected, then ask it to list your pipelines.
Cursor (remote, OAuth)
- 1
Open MCP settings
In Cursor, go to Settings > MCP and choose to add a new server (or edit ~/.cursor/mcp.json).
- 2
Add the Buildkite server
Add an entry with the URL https://mcp.buildkite.com/mcp using the HTTP/streamable transport.
- 3
Complete OAuth and test
Reload Cursor; when prompted, authorize Buildkite in the browser. Then ask the agent to list builds for one of your pipelines to confirm.
Security & permissions
The recommended remote server uses OAuth: after you authorize Buildkite, the server issues short-lived access tokens (valid ~12 hours) with 7-day refresh tokens, so no long-lived API key is stored. The agent's access is bounded by the scopes you grant during the OAuth flow and the permissions of the authorizing Buildkite user. For least privilege, use the read-only endpoint (https://mcp.buildkite.com/mcp/readonly) when an agent only needs to query data rather than trigger or cancel builds. The /direct endpoint and the self-hosted local server instead use a Buildkite API access token passed as a bearer token — keep that token's scopes minimal.
Using Buildkite MCP with Gamut
In Gamut, the Buildkite MCP server becomes a tool your automated agents can call inside event- or schedule-triggered workflows. Gamut manages the OAuth connection to https://mcp.buildkite.com/mcp, so a scheduled agent can, for example, check overnight build status, search failed-job logs, surface failing Test Engine runs, and post a summary — or, on a webhook trigger, retry a flaky job or trigger a build — without anyone wiring up tokens by hand. Scope the connection to read-only where you only need reporting.
Frequently asked questions
What is the Buildkite MCP server?
It is the official Model Context Protocol server from Buildkite that exposes your CI/CD data — pipelines, builds, jobs, logs, artifacts, and Test Engine results — to AI agents and editors. Agents can both read this data and take actions like triggering, canceling, or retrying builds.
Is the Buildkite MCP server free?
Yes. The server is open source under the MIT license and free to use. You do need a Buildkite account, and standard Buildkite plan pricing applies to the underlying CI/CD platform.
How do I connect the Buildkite MCP server to Claude or Cursor?
Point your client at the hosted remote server URL https://mcp.buildkite.com/mcp over the HTTP transport. In Claude Code, run claude mcp add --transport http buildkite https://mcp.buildkite.com/mcp; in Cursor, add the same URL under Settings > MCP. Both complete authentication via an OAuth browser flow on first use.
Is the Buildkite MCP server hosted or self-hosted?
Both. Buildkite hosts a remote server at mcp.buildkite.com (with OAuth, plus read-only and API-token pass-through variants), and you can also self-host the open-source server locally using a Buildkite API access token — the docs recommend running it in a container.