diff --git a/bin/ci/pipeline_message.py b/bin/ci/pipeline_message.py
new file mode 100755
index 00000000000..e4fade392d2
--- /dev/null
+++ b/bin/ci/pipeline_message.py
@@ -0,0 +1,360 @@
+#!/usr/bin/env python3
+# SPDX-License-Identifier: MIT
+
+# Provide a markdown-formatted message summarizing the reasons why a pipeline failed.
+# Marge bot can use this script to provide more helpful comments when CI fails.
+# Example for running locally:
+# ./bin/ci/pipeline_message.sh --project-id 176 --pipeline-id 1310098
+
+
+import argparse
+import asyncio
+import logging
+from typing import Any
+
+import aiohttp
+
+PER_PAGE: int = 6000
+
+
+async def get_pipeline_status(
+    session: aiohttp.ClientSession, project_id: str, pipeline_id: str
+):
+    url = f"https://gitlab.freedesktop.org/api/v4/projects/{project_id}/pipelines/{pipeline_id}"
+    logging.info(f"Fetching pipeline status from {url}")
+    async with session.get(url) as response:
+        response.raise_for_status()
+        pipeline_details = await response.json()
+    return pipeline_details.get("status")
+
+
+async def get_jobs_for_pipeline(
+    session: aiohttp.ClientSession, project_id: str, pipeline_id: str
+):
+    url = f"https://gitlab.freedesktop.org/api/v4/projects/{project_id}/pipelines/{pipeline_id}/jobs"
+    logging.info(url)
+    jobs = []
+    params = {"per_page": PER_PAGE}
+    async with session.get(url, params=params) as response:
+        response.raise_for_status()
+        jobs = await response.json()
+    return jobs
+
+
+def get_problem_jobs(jobs: list[dict[str, Any]]):
+    ignore_stage_list = [
+        "postmerge",
+        "performance",
+    ]
+    problem_jobs = []
+    for job in jobs:
+        if any(ignore.lower() in job["stage"] for ignore in ignore_stage_list):
+            continue
+        if job["status"] in {"failed", "canceled"}:
+            problem_jobs.append(job)
+    return problem_jobs
+
+
+def unexpected_improvements(failed_test_array):
+    if failed_test_array["unexpected_improvements"]:
+        unexpected_improvements_count = len(
+            failed_test_array["unexpected_improvements"]
+        )
+        return f" {unexpected_improvements_count} improved test{'s' if unexpected_improvements_count != 1 else ''}"
+    return ""
+
+
+def fails(failed_test_array):
+    if failed_test_array["fails"]:
+        fails_count = len(failed_test_array["fails"])
+        return f" {fails_count} failed test{'s' if fails_count != 1 else ''}"
+    return ""
+
+
+def crashes(failed_test_array):
+    if failed_test_array["crashes"]:
+        crash_count = len(failed_test_array["crashes"])
+        return f" {crash_count} crashed test{'s' if crash_count != 1 else ''}"
+    return ""
+
+
+def get_failed_test_details(failed_test_array):
+    message = ""
+    max_tests_to_display = 5
+
+    if failed_test_array["unexpected_improvements"]:
+        for i, test in enumerate(failed_test_array["unexpected_improvements"]):
+            if i > max_tests_to_display:
+                message += "  \nand more...<br>"
+                break
+            message += f"{test}<br>"
+
+    if failed_test_array["fails"]:
+        for i, test in enumerate(failed_test_array["fails"]):
+            if i > max_tests_to_display:
+                message += "  \nand more...<br>"
+                break
+            message += f"{test}<br>"
+
+    if failed_test_array["crashes"]:
+        for i, test in enumerate(failed_test_array["crashes"]):
+            if i > max_tests_to_display:
+                message += "  \nand more...<br>"
+                break
+            message += f"{test}<br>"
+
+    return message
+
+
+def get_failed_test_summary_message(failed_test_array):
+    summary_msg = "<summary>"
+    summary_msg += unexpected_improvements(failed_test_array)
+    summary_msg += fails(failed_test_array)
+    summary_msg += crashes(failed_test_array)
+    summary_msg += "</summary>"
+    return summary_msg
+
+
+def sort_failed_tests_by_status(failures_csv):
+    failed_test_array = {
+        "unexpected_improvements": [],
+        "fails": [],
+        "crashes": [],
+        "timeouts": [],
+    }
+
+    for test in failures_csv.splitlines():
+        if "UnexpectedImprovement" in test:
+            failed_test_array["unexpected_improvements"].append(test)
+        elif "Fail" in test:
+            failed_test_array["fails"].append(test)
+        elif "Crash" in test:
+            failed_test_array["crashes"].append(test)
+        elif "Timeout" in test:
+            failed_test_array["timeouts"].append(test)
+
+    return failed_test_array
+
+
+async def get_failures_csv(session, project_id, job):
+    job_id = job["id"]
+    url = f"https://gitlab.freedesktop.org/api/v4/projects/{project_id}/jobs/{job_id}/artifacts/results/failures.csv"
+    async with session.get(url) as response:
+        if response.status == 200:
+            text = await response.text()
+            return text
+        else:
+            logging.debug(f"No response from: {url}")
+            return ""
+
+
+async def get_test_failures(session, project_id, job):
+    failures_csv = await get_failures_csv(session, project_id, job)
+    if not failures_csv:
+        return ""
+
+    # If just one test failed, don't bother with more complicated sorting
+    lines = failures_csv.splitlines()
+    if len(lines) == 1:
+        return ": " + lines[0] + "<br>"
+
+    failed_test_array = sort_failed_tests_by_status(failures_csv)
+    failures_msg = "<details>"
+    failures_msg += get_failed_test_summary_message(failed_test_array)
+    failures_msg += get_failed_test_details(failed_test_array)
+    failures_msg += "</details>"
+
+    return failures_msg
+
+
+async def get_trace_failures(session, project_id, job):
+    project_json = await get_project_json(session, project_id)
+    path = project_json.get("path", "")
+    if not path:
+        return ""
+
+    job_id = job["id"]
+    url = f"https://mesa.pages.freedesktop.org/-/{path}/-/jobs/{job_id}/artifacts/results/summary/problems.html"
+    async with session.get(url) as response:
+        if response.status == 200:
+            return url
+        else:
+            logging.debug(f"No response from: {url}")
+            return ""
+
+
+async def get_project_json(session, project_id):
+    url_project_id = f"https://gitlab.freedesktop.org/api/v4/projects/{project_id}"
+    async with session.get(url_project_id) as response:
+        if response.status == 200:
+            return await response.json()
+        else:
+            logging.debug(f"No response from: {url_project_id}")
+            return ""
+
+
+async def get_job_log(session: aiohttp.ClientSession, project_id: str, job_id: int):
+    project_json = await get_project_json(session, project_id)
+    path_with_namespace = project_json.get("path_with_namespace", "")
+    if not path_with_namespace:
+        return ""
+
+    url_job_log = (
+        f"https://gitlab.freedesktop.org/{path_with_namespace}/-/jobs/{job_id}/raw"
+    )
+    async with session.get(url_job_log) as response:
+        if response.status == 200:
+            return await response.text()
+        else:
+            logging.debug(f"No response from job log: {url_job_log}")
+            return ""
+
+
+async def search_job_log_for_errors(session, project_id, job):
+    log_error_message = ""
+
+    # Bypass these generic error messages in hopes of finding a more specific error.
+    # The entries are case insensitive. Keep them in alphabetical order and don't
+    # forget to add a comma after each entry
+    ignore_list = [
+        "aborting",
+        "error_msg      : None",
+        "error_type     : None",
+        "exit code",
+        "exit status",
+        "exiting now",
+        "job failed",
+        "no files to upload",
+        "ret code",
+        "retry",
+        "retry-all-errors",
+        "unknown-section",
+    ]
+    job_log = await get_job_log(session, project_id, job["id"])
+
+    for line in reversed(job_log.splitlines()):
+        if "error" in line.lower():
+            if any(ignore.lower() in line.lower() for ignore in ignore_list):
+                continue
+            # remove date and formatting before error message
+            log_error_message = line[line.lower().find("error") :]
+            # if there is no further info after the word error then it's not helpful
+            if log_error_message.lower() == "error":
+                continue
+            if log_error_message.lower() == "errors":
+                continue
+            break
+
+        # timeout msg from .gitlab-ci/lava/lava_job_submitter.py
+        if "expected to take at least" in line.lower():
+            log_error_message = line
+            break
+
+    return log_error_message
+
+
+async def process_single_job(session, project_id, job):
+    job_url = job.get("web_url", "")
+    if not job_url:
+        logging.info(f"Job {job['name']} is missing a web_url")
+
+    job_name = job.get("name", "Unnamed Job")
+    message = f"[{job_name}]({job_url})"
+
+    # if a job times out it's cancelled, so worth mentioning here
+    if job["status"] == "canceled":
+        return f"{message}: canceled<br>"
+
+    # if it's not a script failure then all we can do is give the gitlab assigned reason
+    if job["failure_reason"] != "script_failure":
+        return f"{message}: {job['failure_reason']}<br>"
+
+    test_failures = await get_test_failures(session, project_id, job)
+    if test_failures:
+        return f"{message}{test_failures}"
+
+    trace_failures = await get_trace_failures(session, project_id, job)
+    if trace_failures:
+        return f"{message}: has a [trace failure]({trace_failures})<br>"
+
+    log_error_message = await search_job_log_for_errors(session, project_id, job)
+    if log_error_message:
+        return f"{message}: {log_error_message}<br>"
+
+    return message
+
+
+async def process_job_with_limit(session, project_id, job):
+    # Use at most 10 concurrent tasks
+    semaphore = asyncio.Semaphore(10)
+    async with semaphore:
+        return await process_single_job(session, project_id, job)
+
+
+async def process_problem_jobs(session, project_id, problem_jobs):
+
+    problem_jobs_count = len(problem_jobs)
+
+    if problem_jobs_count == 1:
+        message = f"<br>There were problems with job: "
+        message += await process_single_job(session, project_id, problem_jobs[0])
+        return message
+
+    message = f"<details>"
+    message += f"<summary>"
+    message += f"There were problems with {problem_jobs_count} jobs: "
+    message += "</summary>"
+
+    tasks = [process_job_with_limit(session, project_id, job) for job in problem_jobs]
+
+    results = await asyncio.gather(*tasks)
+
+    for result in results:
+        message += result
+
+    message += f"</details>"
+
+    return message
+
+
+async def main(pipeline_id: str, project_id: str = "176") -> str:
+
+    message = ""
+    timeout = aiohttp.ClientTimeout(total=120)
+    logging.basicConfig(level=logging.INFO)
+
+    try:
+        async with aiohttp.ClientSession(timeout=timeout) as session:
+            pipeline_status = await get_pipeline_status(
+                session, project_id, pipeline_id
+            )
+            logging.debug(f"Pipeline status: {pipeline_status}")
+            if pipeline_status != "failed":
+                return message
+
+            jobs = await get_jobs_for_pipeline(session, project_id, pipeline_id)
+            problem_jobs = get_problem_jobs(jobs)
+
+            if len(problem_jobs) == 0:
+                return message
+
+            message = await process_problem_jobs(session, project_id, problem_jobs)
+    except Exception as e:
+        logging.error(f"An error occurred: {e}")
+        return ""
+
+    return message
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Fetch GitLab pipeline details")
+    parser.add_argument(
+        "--project-id", default="176", help="Project ID (default: 176 i.e. mesa/mesa)"
+    )
+    parser.add_argument("--pipeline-id", required=True, help="Pipeline ID")
+
+    args = parser.parse_args()
+
+    message = asyncio.run(main(args.pipeline_id, args.project_id))
+
+    print(message)
diff --git a/bin/ci/pipeline_message.sh b/bin/ci/pipeline_message.sh
new file mode 100755
index 00000000000..84c2825488a
--- /dev/null
+++ b/bin/ci/pipeline_message.sh
@@ -0,0 +1,10 @@
+#!/usr/bin/env bash
+set -eu
+
+this_dir=$(dirname -- "$(readlink -f -- "${BASH_SOURCE[0]}")")
+readonly this_dir
+
+exec \
+  "$this_dir/../python-venv.sh" \
+  "$this_dir/requirements.txt" \
+  "$this_dir/pipeline_message.py" "$@"