acm-jenkins-client

name	acm-jenkins-client
description	Interface to Jenkins CI for reading build status, test results, pipeline stages, console logs, downstream trees, and (with approval) triggering builds. Use when you need Jenkins pipeline data, logs, JUnit results, or to monitor or trigger jobs. Prefer the jenkins MCP in Cursor and Claude Code when configured (same MCP set in both places). Bundled REST docs and scripts/jenkins_api.py cover portability, stdlib-only scripts, and any host without the MCP.
compatibility	Requires VPN for Red Hat Jenkins and API credentials in ~/.jenkins/config.json (jenkins_url, jenkins_user, jenkins_token). In a typical dual setup, the jenkins MCP is configured and is the primary path; references/jenkins-remote-api.md and scripts/jenkins_api.py are the REST fallback and script parity layer (not a substitute for lack of MCP in your env).
metadata	{"author":"acm-qe","version":"1.1.1"}

Jenkins CI Client

Primary path: use the jenkins MCP tools below (expected when Cursor and Claude Code are onboarded with the same MCP stack). That is not a bottleneck for normal QE use.

Also documented: Jenkins Remote API and jenkins_api.py — for stdlib-only automation (e.g. inventory refresh), curl/CI jobs, or any context where the agent does not expose the MCP. See references/jenkins-remote-api.md and python3 "${CLAUDE_SKILL_DIR}/scripts/jenkins_api.py" --help.

This skill exposes how to call Jenkins; the calling skill decides what to extract and analyze.

Prerequisites

• VPN when Jenkins is internal
• Credentials file: ~/.jenkins/config.json (or pass --config to the bundled script) with jenkins_url, jenkins_user, jenkins_token (aliases url, user, token supported)
• Jenkins MCP: use it by default in Cursor and Claude Code when configured; for a few capabilities (rich downstream tree, packaged analyzers) the MCP is still the best interface even though REST covers most reads

Progressive disclosure

• REST / curl / polling / triggers: references/jenkins-remote-api.md
• MCP response field shapes: references/tool-reference.md

Bundled script (no extra pip packages)

Command	Purpose
jenkins_api.py api /job/.../api/json	GET JSON (pretty-printed)
jenkins_api.py console CI-Jobs/foo 123 --tail 500	Last lines of console
jenkins_api.py crumb	Print crumb header for manual curl POST
jenkins_api.py poll CI-Jobs/foo 123	Wait until building is false

Example:

python3 "${CLAUDE_SKILL_DIR}/scripts/jenkins_api.py" api "/job/CI-Jobs/job/ocp_deploy_and_acm_install/lastSuccessfulBuild/api/json?tree=number,url,result"

MCP Tools

Build Information

Tool	Purpose
get_build(job_name, build_number)	Get full build details (result, parameters, duration, timestamp)
get_build_status(job_name, build_number)	Quick build result check (SUCCESS, FAILURE, UNSTABLE, ABORTED)
get_build_log(job_name, build_number, start, max_lines)	Download console log (supports pagination)
get_job(job_name)	Get job configuration and recent builds
get_all_jobs(folder)	List all jobs in a folder

Pipeline Analysis

Tool	Purpose
get_pipeline_stages(job_name, build_number)	Get pipeline stage breakdown with status and duration
get_downstream_tree(job_name, build_number)	Get downstream build tree (parent -> child relationships)
analyze_pipeline(job_name, build_number)	AI-assisted pipeline analysis with failure identification

Test Results

Tool	Purpose
get_test_results(job_name, build_number)	Get JUnit test results (pass/fail/skip counts, per-test details)
analyze_test_results(job_name, build_number)	AI-assisted test result analysis with failure categorization

Build Triggering

Tool	Purpose
trigger_build(job_name, parameters)	Trigger a new build with parameters (use with caution)

Gotchas

1. Job names include folder paths. Use the full path: folder/subfolder/job-name, not just job-name. Same rule for jenkins_api.py console / poll.
2. Build logs can be very large. Use MCP pagination or jenkins_api.py console ... --tail 500.
3. Pipeline stages may be nested. get_pipeline_stages returns a flat list -- parent-child relationships are in the id and parentId fields. REST: try wfapi/describe (see remote API doc).
4. Test results follow JUnit format. failedTests array contains name, className, errorDetails, errorStackTrace. REST: testReport/api/json when the job publishes JUnit.
5. VPN required for typical Red Hat Jenkins hosts.
6. No universal REST substitute for get_downstream_tree / packaged analyzers — use MCP in your normal Cursor + Claude Code setup; REST is for scripts and edge hosts.

Common Patterns

Get a build's test failures (MCP)

get_build(job_name, build_number)          -- build metadata
get_test_results(job_name, build_number)   -- JUnit results with per-test details

Get console log errors (MCP or script)

get_build_log(job_name, build_number, start=-500)  -- last 500 lines

python3 "${CLAUDE_SKILL_DIR}/scripts/jenkins_api.py" console CI-Jobs/your_job 42 --tail 500

Trace downstream builds

get_downstream_tree(job_name, build_number)  -- full downstream tree (MCP preferred)

Rules

• NEVER trigger builds without explicit user approval.
• If the MCP is unavailable or you need a headless/scripted call, use references/jenkins-remote-api.md and jenkins_api.py for reads and polling; for triggers without MCP, print parameters and use UI or documented POST + crumb.