MCP Human-in-the-Loop UI

The goggles — they do
everything.

Give your AI agents rich browser panels for confirmations, forms, dashboards, and diffs — with 9 layers of security built in.

$ pipx install openwebgoggles
View on GitHub Quick Start
Features

Everything agents need to talk to humans

A single MCP server that turns any agent into a full-stack app with browser-based human oversight.

UI

Rich UI Panels

Present decisions to humans with full-fidelity browser UI — not just plain text in a terminal.

Forms Tables Dashboards Diffs Charts
AI

Any Agent Framework

Works with every major agent framework via the Model Context Protocol. One tool, universal compatibility.

LangChain CrewAI AutoGen Agents SDK Claude
9x

9-Layer Security

Defense in depth from localhost binding to Ed25519 signatures. SecurityGate rejects — never mutates.

Ed25519 HMAC CSP SecurityGate Rate Limiting
How It Works

Agent asks, human decides

OpenWebGoggles sits between your agent and the browser, presenting rich UI for human-in-the-loop decisions.

1

Agent

Calls an MCP tool to request a human decision

→
2

OpenWebGoggles

Renders a rich UI panel in the browser

→
3

Browser

Human reviews, edits, and submits their decision

→
4

Response

Decision flows back to the agent as structured data

Developer Experience

Two lines of Python

Import, call, done. The browser panel appears automatically. 

from openwebgoggles.sdk import confirm

# A rich confirmation dialog appears in the browser
approved = confirm("Deploy to production?")

if approved:
    deploy()
Quick Start

Up and running in 60 seconds

Three commands. No configuration files, no Docker, no API keys.

Step 1 — Install

Get the package

pipx install openwebgoggles
Step 2 — Init

Add to your MCP config

openwebgoggles init
Step 3 — Use

Start your agent

claude # or any MCP-compatible agent