One List to Rule Them All: Meet lst.so

lst.so started the way most tools start. Out of frustration.

Not the dramatic kind. The slow, compounding kind. The kind where you realize one morning that you have tasks in five different places and none of them talk to each other.

TL;DR: lst.so is an agent-first task manager that gives you and your AI agents one shared list. Connect Claude Code, ChatGPT, Cursor, or any agent via MCP or REST API. Every task, every subtask, every action logged in one place. $9/month, 7-day free trial at lst.so.

The todo list problem

For years, the system was simple enough. Apple Notes for quick captures. Things for recurring stuff. Project markdown files for dev work. It worked when the only person doing the work was me.

Then AI agents entered the picture.

Claude Code started writing features. ChatGPT handled research and drafts. Cursor refactored code. Suddenly the work was getting done faster than ever, but the tracking? Total chaos.

My personal todos lived in Things. Agent work lived in chat sessions that disappeared. Project tasks lived in .md files scattered across repos. Each agent had its own context, its own memory, its own idea of what was done and what wasn't.

I'd finish a session with Claude Code, close the terminal, and have no record of what just happened. Then I'd open ChatGPT, ask it to pick up where Claude left off, and spend ten minutes re-explaining the state of things. Every. Single. Time.

Five lists, zero visibility

Here's what a typical day looked like:

1. Things had "write blog post" and "review invoices"
2. Apple Notes had a scratch list from a phone call
3. project.md in the repo had "refactor auth middleware" and "add rate limiting"
4. Claude Code session had three subtasks it created on its own that I couldn't see outside the chat
5. ChatGPT thread had research notes and a draft outline that lived nowhere else

Five sources of truth means zero sources of truth. I was the integration layer, manually syncing between all of them. That's not a system. That's a full-time job.

ClawDeck was the first attempt

I built ClawDeck as an open-source kanban board for managing AI agent work. It was good for visualizing tasks, but it was missing the critical piece: agents couldn't interact with the board.

ClawDeck was still a human tool. You'd look at it, drag cards around, update statuses manually. Agents could connect to it via API, but it was still just a kanban board. Great for visualizing work, not great as the single source of truth that replaces everything else.

What I actually needed

One list. Literally just one list where:

• My tasks live next to agent tasks
• Agents can read, create, and complete tasks on their own
• Every action gets logged automatically so nothing disappears
• It works with any agent through MCP or a simple API
• No project management bloat. No Gantt charts. No sprint planning.

That last point matters. Every time I tried an existing tool, it came with ten features I didn't need and was missing the one I did: treating an AI agent as a first-class teammate. Linear and Jira are built for human engineering teams. Notion is flexible but has no concept of an agent picking up a task via API and logging its own progress. None of them let you assign a task to Claude Code and have it autonomously pick it up, work on it, and report back.

So I built lst.so

lst.so is intentionally minimal. Tasks, subtasks, tags, an activity log, and two ways for agents to connect.

MCP (Model Context Protocol) is the emerging standard for AI-tool communication. Install the npm package, add your API key, and your agent discovers all available tools automatically:

{
  "mcpServers": {
    "lst": {
      "command": "npx",
      "args": ["-y", "lst-mcp"],
      "env": {
        "LST_API_KEY": "your-api-key"
      }
    }
  }
}

REST API for custom integrations, scripts, CI/CD hooks, or any agent that speaks HTTP.

Both give full access to tasks, subtasks, tags, and the activity log.

What changed

The morning routine used to be: open Things, check Apple Notes, scan three repo markdown files, scroll back through yesterday's Claude Code session to remember what it did.

Now it's: open lst.so. Everything is there. My tasks, agent tasks, what got done overnight, what's overdue, what's in progress. One screen.

When Claude Code finishes a task, it logs what it did and checks off the subtasks. When I create a task from my phone, agents can see it next time they connect. When ChatGPT does research, the output goes into the task notes, not into a chat thread that'll scroll away.

The five lists collapsed into one.

How it works day to day

1. Create a task: "Refactor the payments controller"
2. Assign it to Claude Code
3. Claude Code picks it up via MCP, reads the codebase, makes changes, and logs every step
4. Open lst.so and the activity log shows exactly what happened

Every task has an owner, you or an agent. Every action gets logged. Every subtask gets tracked. You never have to ask "wait, what did that agent do?"

What makes it different

Agent-first. This isn't a human task manager with an API bolted on. The agent experience is the product. Tasks have an agent field. The activity log tracks who did what, human or AI. Real-time updates mean you see agent progress as it happens.

Tags, not projects. Lightweight, colorful, fast. Filter by tag, agent, status, or due date. No heavyweight project structures.

Recurring tasks. Daily standups. Weekly reviews. Monthly invoicing. Set it once.

One list. Not "one list for work and one list for personal and one list for agents." Just one list. That's the whole point.

FAQ

What is lst.so?
lst.so is a task manager built for people who work with AI agents. It gives you and your agents one shared list with full MCP and REST API access. Agents can create tasks, log progress, check off subtasks, and report back, all without you having to copy-paste between tools.

Does lst.so work with Claude Code?
Yes. Install the lst-mcp npm package, add your API key, and Claude Code can read your tasks, create new ones, log progress, and mark things done. The same setup works for ChatGPT, Cursor, and any MCP-compatible agent.

How is lst.so different from Linear or Jira?
Linear and Jira are built for human engineering teams. lst.so is built for humans who delegate work to AI agents. Every task has an agent assignment field, every action is logged with who did it (human or AI), and agents connect directly via MCP or API to pick up and complete work autonomously.

What does lst.so cost?
$9/month with a 7-day free trial. One plan, full access to everything: unlimited tasks, all agent integrations, MCP server, REST API, and activity logs.

Try it

lst.so is live at lst.so. If you are working with AI agents and tired of tracking tasks across five different places, give it a shot.

I'm building in public and shipping fast. More posts coming soon.

Max