The problem
Most of my working day in Salesforce Marketing Cloud is spent answering small questions. Which automation writes to this data extension? What did that query return last Tuesday? Why did the 6 AM run fail? Each answer sits in a different SFMC app, and by the time four tabs are open I have forgotten why I started.
The Toolkit began life as SFMC Query Saver, a small extension that kept a copy of every query run in Query Studio. Users kept asking for more, so version 2.0 merged Query Saver and my SFMC Automation Viewer into one app. It shipped as an update to the existing store listing with the same storage keys, so nobody lost saved queries in the move.
What it does
- Run Query: up to six SQL panels with lint, autocomplete and run-to-data-extension. When a run fails it shows the real SFMC error instead of a generic one.
- Automations and journeys: activities, run history, watches with a toolbar badge, and journeys drawn like the Journey Builder canvas with a "is this contact in it?" check.
- Data extensions: a "Where used" map, and Create DE from CSV, JSON, Parquet or Excel.
- Email versions: every Content Builder save is kept as a version you can preview and revert to.
- Setup checks: SPF, DKIM and DMARC explained per sending domain, plus alerts before an installed package secret expires.
- On-device AI only: anything AI-assisted runs through Chrome's built-in Prompt API, so no SFMC data leaves the browser for a model.
How it is built
Every call goes through the extension's service worker and reuses the SFMC session you already have in the browser.
The extension never asks for or stores credentials.
It mostly talks to the same internal REST endpoints the SFMC apps themselves use, with SOAP limited to a read-only allowlist of objects.
Data lives in chrome.storage.local and IndexedDB caches on your machine.
Almost everything is read-only. The first feature that writes to SFMC, removing contacts from a journey, makes you type the number of contacts to confirm, and an admin policy can switch it off. A CI guard fails the build on any change to permissions or the content security policy, because a new permission disables the extension for every user until they accept it.
What I learned
- A standalone Query Activity only reports "is running". To surface the real error, the Toolkit re-runs the activity once inside a temporary automation, reads the run log, then deletes the automation.
- A copied primary key silently drops rows. A
UNION ALLof 40 rows twice returned 40, so result data extensions are now created without a primary key. - Automation Studio's legacy
$skipis a page number, not a row offset. Lists quietly stopped at 100 until a user reported it. - After a business unit switch, Content Builder and Journey Builder can stay in the old unit. The fix compares org IDs before each call.
- Test doubles lie. Chrome stored a reactive Vue array as an object, and the fake storage in tests hid it completely.