AI Copilot Is Now a Must-Have, Here's An Anecdote

Harshal
1 day ago
7 min read

Why I Moved from Zoho Projects to Trello

I started losing tasks at the moment I needed to capture them. A task would pop into my head. It was often a small house task or a small personal task. I had maybe 30 seconds. If I opened a project management tool, I would click around, lose momentum, and postpone. Sometimes I would forget the task.

Trello MCP changed that capture moment for me. This experience made one thing clear to me. AI copilots are now a must-have feature for products to be successful. If a tool does not support an AI copilot well, that tool feels slower and harder to use.

I heard about Trello MCP on the How I AI podcast, where Teresa Torres was the guest. I wanted to try it because I was wasting a lot of time opening different task boards and updating them. My goal was to give an AI agent a command and have the agent handle most board updates.

Most of the time, I use voice (Wispr Flow) with Cursor. That workflow works well. I can dictate the task to the agent to create or update the card in Trello. This is why I moved from Zoho Projects to Trello.

I'll share the different trials I did and learnings.

You need 7 minutes to read this.

simple Kanban board, glowing with an AI copilot

Problem context: my multi-board reality

I use multiple boards:

With my wife, for house tasks like paperwork and smart home fixes
With freelancers, for data analysis, programming, and automations
With a personal assistant, for data entry and web research tasks

This setup enables me to not drop the ball or coordinate with others. But if I am slow at making small updates, the backlog gets stale and I lose time.

I noticed this impact:

Cards are more likely to have labels and more context.
I update cards more often, especially small edits like adding labels or extra context.

Takeaways From Trials

Two things stood out from this migration:

AI changes product usability. Trello and Zoho Projects do not have built-in AI, but I could connect one better than the other. The one I could connect became faster and easier to use.
Test at a small scale before writing a bulk script. An AI coding agent can generate a complete Python script immediately, but I needed to understand the process first. Migrating one task with MCP helped me give better context to the agent before migrating all tasks.
No plan means failure. Building without a clear plan, a concrete end goal, or an easy way to test almost always leads to problems, even if you are exploring or experimenting. For example, finding the API spec for a product you want to integrate with.
AI builders need feedback. AI builder agents work best when they can test and debug what they build. Make feedback and testing easy, so your AI builder can see the results and improve its output. For example, a disconnected coding agent cannot test the Chrome extension it is building.

Approaches considered

Zoho built-in MCP with Cursor
Trello community MCP (Cursor, Codex, AMP)
Zoho in n8n
Trello in n8n (free MCP template)
Trello in n8n (paid template)
Zapier MCP for Trello
Chrome extension as Zoho co-pilot
Migrate to Trello and use Trello MCP

Zoho Projects with MCP

I tried to build an n8n workflow to act as a co-pilot for Zoho Projects, but with no existing nodes, I would need to create HTTP Request Nodes for each resource and operation.

Zoho has a built-in MCP

I used the Zoho built-in MCP with Cursor. It made Cursor crash. There were too many tools exposed by the MCP. My guess is the tool descriptions were too long but still not clear. Cursor would often forget how to use the MCP or the MCP ran into authentication issues.

I also tried it with n8n, but that didn't work well either.

Starter Zoho Projects Agent using n8n and Zoho MCP

Zoho Projects Chrome extension

Next, I built a Chrome extension using Cursor to work as an AI co-pilot for Zoho Projects. But, my mistake was in not reviewing the plans or providing precise spec with API endpoint specific details. My hope was that whenever I have Zoho Projects open, I can open this sidebar and it would act as an AI assistant. I got this idea because I saw some such community products built for n8n. I built this extension by telling the coding agent of Cursor to look up the API specification of Zoho projects and implement a copilot for it.

Zoho Projects Agent using Chrome extension

I found it very helpful to also think in terms of enabling users to customize settings instead of hard coding anything. I put my built-in public hat on to think of if I need to publish this code on GitHub, then what would keep it secure? A lot of these are standard development practices, and I realize that the more I built a project, the more likely I was to come back and follow standard development practices.

My AI co-pilot did not work well.

Settings customization for Chrome Extension

Trello MCP in n8n

I could build an n8n workflow using the Manage Trello tasks with AI assistants via MCP server template that I could trigger from my phone or laptop, anywhere, to update my board. n8n had built-in Trello nodes, but they didn't cover all the features I wanted, especially getting all the boards and lists. It expected me to know my board ID and list ID. I could solve this using HTTP request nodes. The Trello MCP template worked well for straightforward tasks written in very clear, precise language but could not handle even a little bit of ambiguity or just a difference in phrasing of the request.

I triggered my n8n workflow from Telegram. I could input images, audio, documents, or text. It used the MCP to respond back to me.

n8n workflow (left) using Trello MCP template (right)

I found a paid n8n template that worked with Trello: Trello tool MCP server (all 41 operations) via n8n It had several tools so it was more comprehensive, but because it was paid, I shelved it to try other build-your-own approaches first. A potential next step for me would have been to give a screenshot of this paid template and an example smaller Trello MCP workflow to Cursor and ask it to re-create it.

I also noticed a Zapier Trello MCP, but I did not try it.

Trello Community MCP with Cursor

I found two community MCPs for Trello and chose the one recommended by Perplexity. This MCP had several tools so it was more comprehensive, but because it was community, I was not sure how reliable it was. Thankfully Teresa Torres had used it successfully, so I went ahead and set up Trello MCP with Cursor, Codex, and AMP. They all worked. And they worked very well. I considered delorenj/mcp-server-trello and kocakli/trello-desktop-mcp.

I've written more on this in Set Up Trello MCP in Cursor To Chat With Your Project Boards.

Trello AI vs Zoho Projects AI

This difference in experience made me rethink how I use Trello and Zoho Projects. My content backlog was in Zoho Projects, but my Household Tasks Backlog and Personal Assistant Board were in Trello. I liked Trello's UI more, but I had premium Zoho Projects through the Zoho One pricing plan. I wanted to stick to one fully paid for tool, instead of using two project management tools at the same time.

At first, the free version of Trello had limits for my setup. I could not build the automations I wanted. I also could not clearly see epics or task priorities. I couldn't sort the cards by priority.

But now,

I noticed that labels show full text when I click them, and remain visible.
I use Comet browser to make changes in Trello.
I discovered the Trello API. So, I could build n8n automations (e.g. auto-archiving old cards).
I can use Trello MCP to chat with my boards, and that could reduce my need to look at an unsorted board.

That solved most of the concerns I had about using Trello's free version.

Migration script using Cursor, MCP, and API

I built a migration Python script with Cursor. The script reads every item in an export from Zoho Projects and uses the Trello API to create missing cards.

Workflow:

Export tasks from Zoho Projects (CSV or similar).
Use Trello MCP in Cursor to get board ID, list IDs, and label IDs.
Write a Python script that reads the export and calls the Trello API.
Run the script (test with a small batch first).

Identified IDs from MCP then used for API

I ran a dry run first with one item to verify all field mappings. For the full transfer, the script had verbose logging that reported how many items it imported. I compared that count to the exported CSV. The script also reported two items it could not import. I checked those in the logs to understand why and confirm completeness.

Task name (card title), description, due date, labels, list placement transferred. Attachments, comments, subtasks, time logs, task dependencies did not transfer.

I first used Cursor MCP to get basic information and understand the API and board structure. Then I used that context to build a deterministic script from CSV to Trello.

This way I could use AI to do the ambiguous parts, which was figuring out the board and a tile list ID, or maybe creating new labels or new lists. I could use the API, which is much cheaper, faster, more reliable, and deterministic, to do the bulk transfer of hundreds of items.

My Takeaway

AI copilots change how usable a product feels. If your tool does not integrate well with one, it will feel slower than alternatives that do. If you’re evaluating tools, test copilot support first. If you are building a product, AI copilot is a must-have feature to be successful.