A plain text plan can coordinate many AI agents with less overhead
Notal is a markdown notebook that also works as an MCP server. A person writes a plan once in plain text, and different AI agents use that same page as their shared work list. Each agent asks for only the tasks that are ready now, chooses one, then reads the prompt, role, and for that task.
The agent does the work in its own environment, such as Cursor, Claude Code, a cron job, or . It then records whether the task succeeded, failed, became blocked, what evidence supports the result, and when the task should run again. The design avoids sending the whole notebook to the agent every time and instead gives it only the current queue.
The main idea is that Claude, Cursor, Codex, Goose, Hermes, LangGraph, or a custom worker can all follow the same plan file instead of being tied to one system.
Key points
- A markdown file acts as both the plan and the shared task queue.
- Agents receive only tasks that are ready now, not the full notebook.
- Each task can include a prompt, role, and .
- Results are logged with success, errors, blocked status, evidence, and the next due time.
- The same plan can be used by many , including Cursor, Claude Code, Codex, and .