OK, I’ve dipped my toe into the Model Context Protocol (MCP) world. In case you didn’t know, MCP is a mechanism that allows an AI to perform work on your behalf.
Before this week, I usually asked my old AI (Gemini) for its opinion and copied + pasted its answer. Sometimes I would let it update code on my local computer, but that’s all.
As of this week, my new AI (Claude) is connected to:
- A Salesforce development org
- Trello (for project management)
- Git (for source control)
My local working environment is VS Code with the Salesforce and Claude Code extensions.
My workflow looks like this:
- Create a Trello card and add details about the feature, update, or bug
- Have the AI read the card and design an approach
- Refine the design through a natural conversation, as needed
- Approve the design
- Have the AI build everything locally
- Review the changes and push them to Salesforce
- Unit test in Salesforce and return to step 3 if needed
- Have the AI update the card with the final design, development notes, and test steps
- Have the AI create or update client documentation
- Have the AI push changes to Git (using a branching strategy)
The takeaway
Using this workflow, I’ve completed 27 cards in the last 4 days. Mic drop.
Category:
Salesforce