Audit the data sources inside a .twbx.

What is the direct answer to this workbook question?

A .twb is XML, and a .twbx is a zip with that XML plus packaged data. Each data source's connection (server, database, user), its tables and joins, any Custom SQL and any extract are written in the <datasources> section. Drop a workbook below to list them in your browser, or run the Python script on this page over a folder of workbooks. No database credentials or Tableau install needed.

When should I audit workbook data sources?

Before a warehouse migration or a dbt refactor, someone has to find which workbooks still read a deprecated table or carry their own Custom SQL. Opening forty workbooks one by one in Tableau Desktop is slow, and the people doing the audit often don't have a Tableau seat. Tableau Desktop does have a free edition for one person: Tableau Desktop Free Edition: full Desktop analytics for one person, stored locally; no publishing, online sharing or sharing packaged workbooks. A Creator seat costs $75 per user/month, billed annually (checked 2026-10-18). For an audit you need neither: everything is in the file.

Which XML nodes describe connections and queries?

Read connection attributes for server and database names, relation nodes for tables and joins, and text relations for Custom SQL.

How can I run the same audit from a script?

Run it on one workbook or a whole folder (python3 audit.py *.twbx), or hand it to a coding agent such as Claude Code or Codex and ask for a lineage table. It uses only the Python standard library. On the sample workbook above it prints the same server, tables, join, Custom SQL, extract and packaged file the page shows.

import sys, zipfile
import xml.etree.ElementTree as ET

def audit(path):
    if path.lower().endswith(".twbx"):
        with zipfile.ZipFile(path) as zf:
            names = zf.namelist()
            twb = next(n for n in names if n.lower().endswith(".twb"))
            root = ET.fromstring(zf.read(twb))
            print("Packaged files:", [n for n in names if n != twb and not n.endswith("/")])
    else:
        root = ET.parse(path).getroot()
    for ds in root.iter("datasource"):
        if ds.get("name") == "Parameters" or ds.find("connection") is None:
            continue
        print(f"\n=== {ds.get('caption') or ds.get('name')} ===")
        for c in ds.iter("connection"):
            if c.get("class") not in (None, "federated", "hyper"):
                print("Connection:", {k: c.get(k) for k in ("class", "server", "dbname", "username", "filename") if c.get(k)})
        for rel in ds.iter("relation"):
            if rel.get("type") == "table":
                print("Table:", rel.get("table"))
            elif rel.get("type") == "join":
                print("Join:", rel.get("join"))
            elif rel.get("type") == "text":
                print("Custom SQL:\n" + (rel.text or "").strip())
        for ex in ds.iter("extract"):
            for c in ex.iter("connection"):
                print("Extract:", c.get("dbname"))

for workbook in sys.argv[1:]:
    print(f"##### {workbook}")
    audit(workbook)

How can I move Custom SQL out of a workbook?

Tableau runs Custom SQL as a subquery, wrapping your statement in its own SELECT, and the logic is invisible to warehouse lineage tools. Once the audit has found it, move the query into a versioned view or dbt model (for example analytics.valid_orders) and point the workbook at that table. The workbook keeps working, and the next migration can see the dependency.

What can a packaged workbook expose?

Tableau's documentation describes a packaged workbook as the workbook plus copies of local file data sources and extracts. Anyone who has the file can read the connection details, filter values and the packaged data, even without access to the database. Check what is inside before you email one.

What data does this worked example use?

The initial view uses authored synthetic data, not customer data or a captured agent session.

The example record lists its file, calculation or SQL and expected values. The live browser result is computed by xmetric’s existing engine or workbook inspector. Samples are excluded from own-data activation reporting.