I initially built this for myself but some folks asked me to open access to it. Let me explain what I built and why.

I'm a product designer who transitioned from only designing to building complete software products myself and I found myself using AI coding agents literally every day. I started with cursor, then codex and claude code, then kilo code, and then moved to opencode.

The only thing AI agents couldn't get done for me was to keep track of my todos and tasks for all my projects. I ended up vibe coding a todo app separately or using apps like TickTick. The main issue with separate apps was that I needed to constantly switch context in the middle of my work sessions and repeat the tasks I needed to get done to the agent over an over.

Then I realized I can just give the agents the full todo list context and tools to do CRUD operations directly in the conversations without the need for a separate app, so that way I can simply prompt my agents to check what's due soon, update tasks, add notes, create new ones, plan my weeks, etc. All in the same conversations as the projects I work on through opencode, which works great for mid-session requests like adding a new subtask based on the current agent's work output.

I built this as a first-class MCP platform that supports the new stateless HTTP configuration as well as the standard SSE, so setup is a single prompt and your agents stay in sync with your actual work state immediately.

It works with any agent harness or app that supports MCPs. Which means that if you use multiple agents, all of them will be in sync with the same tasks whether you use codex first, then switch to claude code, and then to opencode in the same day.

There's also a web-app UI access for doing any updates manually or if you need to use it on the go with your phone. You'll always see what was done by you in the web-app or by your agents through the MCP bridge.

I just launched it on producthunt and you can try it free if you want:

Live app: https://taskshell.app Producthunt: https://www.producthunt.com/products/taskshell

In the web-app there's a "report bug" form if you try it and something doesn't work for you. All feedback is appreciated!

Show HN: I built an agent-first productivity bridge for all your agents