Skip to content

I Cancelled Cursor. Then I Bought It Again.

· 5 min read

I cancelled Cursor, wrote about why, and then bought it again.

That is not because I changed my mind about everything. It’s because I tried to make my Claude Code subscription work properly through OpenCode, and it just did not configure correctly. My OpenCode setup is too central to how I work for me to treat that as a minor inconvenience.

If the setup is wrong, the whole workflow is wrong.

What broke

The idea was simple. Keep the Claude subscription, use it through OpenCode, keep all the custom workflow stuff I already rely on.

That last part is the important one. OpenCode for me is not “the app I happen to like.” It’s the place where all the weird useful parts of my workflow live:

  • custom tooling
  • custom prompts
  • workspace isolation
  • model switching
  • all the MCP glue that makes the whole thing useful

So when the Claude Code subscription did not wire up cleanly there, the problem wasn’t “annoying config.” The problem was that the environment I depend on stopped being trustworthy.

I don’t care how good the model is if I don’t trust the setup around it.

So I bridged Cursor back in

Instead of trying to force the Claude Code route for another week, I built a local MCP bridge back into Cursor.

That let me keep the parts of OpenCode that are critical to how I work while getting Cursor’s coding agents back into the loop.

And honestly, Cursor’s agents are still really good.

That was the thing I kept running into while trying other tools. I could find cheaper options, more configurable options, more open options, better terminal options. But Cursor kept winning on the boring part that matters: I ask it to do code work inside a repo and it usually does that code work well.

Composer got good

Composer is also in a much better place than it was when I last had strong opinions about Cursor.

It feels more usable now, the pricing landed in a place that makes sense, and it no longer feels like a feature I have to defend on principle. I use it because it is good.

That matters more than “this could be powerful if you squint at it.”

Cursor is making the right bet

The biggest reason I feel comfortable going back is that Cursor seems to be heading in the right direction strategically.

They’re not forcing one model worldview on you. I can use basically any model I want. That’s the right move.

I don’t want to rebuild my entire workflow every time one lab ships a better coding model. I don’t want my editor choice to become a pricing or vendor lock-in decision. If a tool gives me the agent layer, editor integration, and model flexibility, I’m interested.

Cursor is doing that.

Cursor Glass makes the direction even clearer

Cursor Glass just came out, and it’s basically Cursor moving further toward the kind of workflow I wanted from OpenCode in the first place.

The important parts are not the branding, they’re the capabilities:

  • custom MCP tools
  • coding agents that can actually do useful work
  • vector search across the codebase
  • sub-agents
  • workspace isolation

That is the stack I care about.

Vector search helps a lot once your codebase gets big enough that exact string matching stops being enough. Sub-agents matter because I don’t want my main agent spending half its context window doing reconnaissance. Workspace isolation is critical because once you start running parallel work, context bleed becomes a real problem.

Those are not “nice to have” features for me. They are workflow features.

Why not just stay on Claude Code

Claude Code is still well built in a lot of ways. The defaults are good. You can get to useful work quickly. If you want the opinionated path and that path matches how you work, it makes sense.

But I don’t want the opinionated path. I want my path.

That usually means custom tools, weird glue code, different models for different tasks, and an environment I can keep reshaping as the tooling changes.

That is also why I still care so much about OpenCode. It’s not because I think every part of it is more polished. It obviously isn’t. It’s because the extensibility is real.

I looked at T3 Code too

I also looked at T3 Code.

I get the appeal. It seems to be going after a lot of the right ideas. But the UI is not really to my liking, and if I am going to spend all day in a tool, that matters. A lot.

People talk about UI like it is superficial. It isn’t. Bad UI adds friction to every single action. Good UI disappears.

Cursor is just better for me there.

So yes, I am back on the $200 Cursor plan

That is where I landed.

I tried to keep the Claude Code subscription path as the core setup. It didn’t wire up correctly in the environment I actually depend on. Cursor’s agents are strong, Composer is good now, Glass pushes further in the right direction, and the model flexibility means I don’t feel boxed in.

So I paid for Cursor again.

For the foreseeable future, that is probably where I’m staying.

Not because it won on ideology. Because it won on workflow.