PlugWiki
buildkite-mcp-server logo

Buildkite MCP, Buildkite's official MCP server

OfficialRemoteLocal

Lets your AI assistant check Buildkite builds, read job logs, diagnose failures, and retry or start builds.

This is Buildkite's own server. It lets Claude, Cursor and other assistants find why a build failed, read job logs and test results, and, if you allow it, retry jobs or start builds. Buildkite hosts a version you sign in to with your Buildkite account, and a separate read-only address cannot make changes. You can also run it yourself on your machine with an API token.

Good for

  • Developers who want to know why a build is red without clicking through logs
  • Platform engineers watching agents, queues and clusters
  • Test owners hunting slow and flaky tests in Test Engine
  • Team leads who want a quick answer on what is blocked

Works in

14 should work

Tap an app to see how to set it up and any known problems.

Install buildkite-mcp-server

Claude Code

  1. 1

    Copy this prompt.

    Add the buildkite-mcp-server server to Claude Code. Run this command:
    
    claude mcp add --transport http buildkite https://mcp.buildkite.com/mcp
    
    Ask me for BUILDKITE_API_TOKEN and only put the values in that config, nowhere else.
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Sign-in with the vendor account (OAuth).

Built from the package data we hold. Not yet tested by us.

Tools

  • Explains why a build failed in one step

    Diagnoses a failed build in one call, with the build state, a count of jobs by state and details of the problem jobs.

    Try: “Why did build 1234 on the web pipeline fail?”

    Network · get_build_failure_summary

  • Lists recent builds, newest first

    Lists builds for one pipeline, or across all pipelines in an organization, with light details.

    Try: “Show the last 10 builds on main for the web pipeline.”

    Network · list_builds

  • Shows one build

    Returns a single build with short annotation summaries.

    Try: “Show me the details of build 1234.”

    Network · get_build

  • Compares a build with an earlier one

    Compares step results, retries and timing between builds of one pipeline. Picks an earlier build in the same pipeline as the baseline.

    Try: “What changed since this build last passed on main?”

    Network · compare_builds

  • Waits until a build finishes

    Polls a build until it passes, fails, is canceled or is blocked.

    Try: “Wait for build 1234 to finish and tell me the result.”

    Network · wait_for_build

  • Lists the jobs in a build

    Lists jobs for a build with a short summary, and can filter to failed or timed-out jobs.

    Try: “List the failed jobs in build 1234.”

    Network · list_jobs

  • Shows one job

    Returns a single job by its ID.

    Try: “Show me details of the failed test job.”

    Network · get_job

  • Shows a job's environment variables

    Returns the environment variables set for a job. This can expose values you would rather keep private.

    Try: “Show the environment variables of the deploy job.”

    Network · get_job_env

  • Shows the last lines of a job log

    Returns the last entries of a job log, which is where most failures show up.

    Try: “Show the last 50 lines of the failed job's log.”

    Network · tail_logs

  • Searches a job log with a pattern

    Searches log entries with a regular expression and optional surrounding lines.

    Try: “Search that log for "ECONNRESET" with 5 lines of context.”

    Network · search_logs

  • Reads a job log from a chosen row

    Reads log entries, optionally starting at a row number, with a limit.

    Try: “Read the first 100 lines of that job's log.”

    Network · read_logs

  • Lists the pipelines in an organization

    Lists pipelines with basic details, build counts and status.

    Try: “List all pipelines in my organization.”

    Network · list_pipelines

Read from the source code

Permissions

medium risk · 4 things to watch

Medium risk

The assistant can start, retry, cancel and unblock builds, and Buildkite builds can run deploy steps. Begin with the read-only address and add write access when you need it.

What it can do

Watch out for

  • It can start and unblock builds that deploy

    Starting a build or approving a block step can run pipeline commands, including deployments. Keep approval prompts on for write actions in your AI tool. buildkite.com

  • It can read job environment variables and logs

    Logs and job settings can contain secrets that your pipeline prints or sets. Everything it reads goes to your AI provider. github.com

  • It can create secrets and change queues

    Tools can store cluster secrets, pause a queue and change pipeline settings. The read-only address removes all of them. github.com

  • Read-only is not a sandbox

    Buildkite notes that an agent with shell access and a signed-in Buildkite CLI can still make changes with the CLI. buildkite.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Turning on read-onlyHosted: use https://mcp.buildkite.com/mcp/readonly, which only offers tools that need read scopes. Local with Docker: add -e BUILDKITE_READ_ONLY=true to the docker run command. Local binary: add the --read-only flag. You can also give the local server a token with only read scopes.
Talks toBuildkite (mcp.buildkite.com when hosted, api.buildkite.com when run locally)
Scopes
  • Hosted server: OAuth with read and write access, or read-only on the read-only address

  • Local server: a Buildkite API token, whose scopes decide which tools work

Questions

What is the Buildkite MCP server URL?

It is https://mcp.buildkite.com/mcp. It signs you in with OAuth and has read and write access. The read-only version is https://mcp.buildkite.com/mcp/readonly.

Can I make it read-only?

Yes. On the hosted server, connect to https://mcp.buildkite.com/mcp/readonly, which only offers tools that need read scopes. On the local Docker server, add -e BUILDKITE_READ_ONLY=true to the docker run command. On the local binary, add the --read-only flag. Read-only limits this connection only, not an agent's shell or other signed-in tools.

Hosted or local: which should I use?

Use the hosted server for your own use in a chat tool, as Buildkite recommends. Run the local server when you need a pinned version or an automated agent that cannot sign in through a browser.

How do I run the local server with Docker?

Run the image buildkite/mcp-server with the stdio subcommand and your token as BUILDKITE_API_TOKEN. In Claude Code: claude mcp add buildkite -- docker run --pull=always -q --rm -i -e BUILDKITE_API_TOKEN=<your-token> buildkite/mcp-server stdio. Buildkite's setup page has the same command for other clients, and a binary route as buildkite-mcp-server stdio. Use buildkite/mcp-server:<version> to pin a version.

What does the API token need?

It depends on the tools you use, and only the local server uses a token. Buildkite's smallest set is the read_builds, read_pipelines and read_user scopes. Adding read_clusters, read_secrets_details, read_artifacts, read_build_logs, read_organizations and read_suites gives all read-only tools, and write tools need more write scopes. A retries-only setup needs read_builds, read_build_logs and retry_builds.

How many tools does it have?

The tool list above is partial. The server has dozens of tools in many groups, and you can limit them to chosen groups when you run it locally.

How do I add it to Claude Code or Cursor?

Use the install box above. Both sign you in to Buildkite the first time.

Information

updated Nov 2, 2026

Publisher
Official
Transport
Streamable HTTP, stdio
Auth
OAuth
Version
0.7.0
License
MIT
Last commit
Oct 31, 2026
Open issues
3
Listed in
Official MCP registry, Docker MCP Catalog

GitHubDockerOfficial MCP RegistryChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 7
  1. Buildkite MCP server docs · buildkite.com
  2. buildkite-mcp-server README · github.com
  3. buildkite-mcp-server main.go · github.com
  4. buildkite-mcp-server stdio.go · github.com
  5. buildkite.com · buildkite.com
  6. buildkite.com · buildkite.com
  7. buildkite.com · buildkite.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.