Buildkite MCP via mcporter + mcp-remote

Use this skill to access Buildkite data through the Buildkite MCP server using mcp-remote as a stdio proxy. This avoids flaky direct OAuth flows.

Setup

•Add the server (home scope)

bash

mcporter config add buildkite-remote \
  --command "npx" \
  --arg "-y" \
  --arg "mcp-remote@latest" \
  --arg "https://mcp.buildkite.com/mcp" \
  --arg "--transport" \
  --arg "http-only" \
  --arg "--auth-timeout" \
  --arg "60" \
  --arg "--host" \
  --arg "127.0.0.1" \
  --description "Buildkite MCP via mcp-remote" \
  --scope home

Use --scope project if you want the server stored in config/mcporter.json for just the current repo.

•Authenticate (browser flow)

bash

mcporter auth buildkite-remote

Keep the CLI open after the browser says “auth succeeded.” The callback hits 127.0.0.1, so the CLI must run on the same machine as the browser.

If you see OAuth error: invalid_request, clear cached state and retry:

bash

mcporter auth buildkite-remote --reset

•Verify auth

bash

mcporter call 'buildkite-remote.access_token()' --output json

•Confirm the server and schema

bash

mcporter list buildkite-remote --schema

Common calls

Fetch a build and list jobs (use function-call syntax so build numbers stay strings):

bash

mcporter call 'buildkite-remote.get_build(org_slug: "ORG", pipeline_slug: "PIPELINE", build_number: "BUILD_NUMBER", detail_level: "full")' --output json

Search a job log (reverse search with context):

bash

mcporter call 'buildkite-remote.search_logs(org_slug: "ORG", pipeline_slug: "PIPELINE", build_number: "BUILD_NUMBER", job_id: "JOB_ID", pattern: "\\bfailed\\b", reverse: true, limit: 80, context: 3)' --output json

Read a focused slice of the log:

bash

mcporter call 'buildkite-remote.read_logs(org_slug: "ORG", pipeline_slug: "PIPELINE", build_number: "BUILD_NUMBER", job_id: "JOB_ID", seek: SEEK, limit: 100)' --output json

Downloading artifacts

•List artifacts for a job and extract the download URL:

bash

mcporter call 'buildkite-remote.list_artifacts_for_job(org_slug: "ORG", pipeline_slug: "PIPELINE", build_number: "BUILD_NUMBER", job_id: "JOB_ID", page: 1, perPage: 100)' --output json \
  | jq -r '.items[] | [.path, .download_url] | @tsv'

•Attempt direct download via MCP (returns base64 data on success):

bash

mcporter call 'buildkite-remote.get_artifact(url: "DOWNLOAD_URL")' --output json \
  | jq -r '.data' \
  | base64 --decode > artifact

•If you see Only one auth mechanism allowed (S3 rejects the Authorization header), extract the presigned S3 URL from the MCP error and download with curl (no auth header):

bash

DOWNLOAD_URL="https://api.buildkite.com/v2/organizations/ORG/pipelines/PIPELINE/builds/BUILD_NUMBER/jobs/JOB_ID/artifacts/ARTIFACT_ID/download"
S3_URL=$(mcporter call "buildkite-remote.get_artifact(url: \"${DOWNLOAD_URL}\")" --output json \
  | perl -ne 'if (/GET (https:\/\/s3.amazonaws.com[^ ]+)/) { $u=$1; $u=~s/:$//; print $u }')

curl -L "$S3_URL" -o artifact

Notes

•Use --output json so you can script the results.
•Adjust --auth-timeout or --transport if the remote flow is slow.