Metadata-Version: 2.4
Name: loopdesk
Version: 0.0.1
Summary: Local control desk for AI agents (MCP server + web UI): tasks with completion criteria, recorded assumptions, one-click human approval and clarification.
Keywords: mcp,model-context-protocol,ai-agents,human-in-the-loop,claude-code
Author: Axel Solhall
License-Expression: MIT
License-File: LICENSE
Classifier: Development Status :: 1 - Planning
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Software Development
Requires-Python: >=3.11
Project-URL: Homepage, https://loopdesk.dev
Project-URL: Repository, https://github.com/axelsolhall/loopdesk
Description-Content-Type: text/markdown

# loopdesk

A local control desk for AI agents: an MCP server your coding agents (Claude Code, Codex, …) connect to, and a web UI in
your browser where you steer them.

> **Status: planning.** Version 0.0.1 holds the name; nothing is usable yet.

## The idea

With capable agents, the limit is no longer how fast code gets written but how fast a human can understand and steer
the work. loopdesk keeps the models busy and keeps the human's part small and quick:

- **Tasks with a finish line.** Agents queue tasks, each with a well-defined goal that says when it is done; the
  result (code, text, images, …) waits for you to accept or reject.
- **Feedback before execution.** Tasks that need your input are put in front of you before they run — as multiple
  choice, a text box, or a button that asks the agent for clarification (shift-click to add your own question).
- **Recorded assumptions.** Every assumption an agent makes is written down, with what depends on it, so a wrong one
  is cheap to find and fix.
- **Work for subagents.** Approved tasks are split into units subagents can take on their own.
- **Search.** A local vector index over tasks, assumptions and decisions.

Runs locally: `pip install loopdesk`, start the server, open it in your browser.

## License

MIT
