Cette page est affichée en anglais. Une traduction française est en cours.
SkillsMay 12, 2026·2 min de lecture

Magentic-UI — Human-Centered Web Agent with MCP

Magentic-UI gives teams a human-in-the-loop web agent with plan previews, approval gates, Docker-backed execution, and MCP support.

Prêt pour agents

Staging sûr pour cet actif

Cet actif est d'abord staged. Le prompt copié demande à l'agent d'inspecter les fichiers staged avant d'activer scripts, config MCP ou config globale.

Stage only · 29/100Policy : staging
Surface agent
Tout agent MCP/CLI
Type
Skill
Installation
Stage only
Confiance
Confiance : Established
Point d'entrée
Asset
Commande de staging sûr
npx -y tokrepo@latest install 2768e070-316f-5eeb-80c8-14d85cb987d8 --target codex

Stage les fichiers d'abord; l'activation exige la revue du README et du plan staged.

Introduction

Magentic-UI gives teams a human-in-the-loop web agent with plan previews, approval gates, Docker-backed execution, and MCP support.

  • Best for: teams that need browser automation with explicit plans and approval steps
  • Works with: Python 3.10+, Docker, web UI, command-line mode, MCP agents, optional Azure or Ollama installs
  • Setup time: 15-30 minutes

Practical Notes

  • Quant: the README supports Python 3.10-3.13 and starts on http://localhost:8081.
  • Quant: it documents at least three runtime modes out of the box: Docker-backed UI, --run-without-docker, and the magentic-cli path.

Rollout pattern

  • Use the default Docker-backed path first so you validate the intended execution environment.
  • Pilot low-risk web tasks such as monitoring or read-only extraction before approving form submissions or purchases.
  • Add MCP agents only after you understand what the core UI agent already does well on its own.

Watchouts

The no-Docker mode is explicitly limited, so teams should not assume it offers the same execution guarantees as the full default setup.

FAQ

Q: Can I run it without Docker? A: Yes, but the README marks --run-without-docker as limited because code execution is unavailable.

Q: Why is this safer than a black-box browser agent? A: It exposes the plan and asks for approval on sensitive operations.

Q: What is the first real test? A: Open the UI, inspect the proposed steps, and approve a low-risk browsing or monitoring task.

🙏

Source et remerciements

Source: https://github.com/microsoft/magentic-ui > License: MIT > GitHub stars: 9,805 · forks: 979

Fil de discussion

Connectez-vous pour rejoindre la discussion.
Aucun commentaire pour l'instant. Soyez le premier à partager votre avis.

Actifs similaires