Vigilance

A Go-based MCP (Model Context Protocol) server for AI agent observability — real-time intent tracking, visual attention monitoring, and behavioral telemetry for AI coding workflows.

What It Does

Vigilance sits between AI agents and their tools, observing what the agent is looking at, what it's planning to do, and whether its behavior matches expectations. It provides:

• Real-time visual frame ingestion — captures and streams viewport frames from Unity or Blender
• MJPEG streaming — low-latency video relay with SVG overlay support for gaze frustums and bounding boxes
• Intent tracking — intercepts MCP traffic to stream the AI's reasoning and planned actions
• Dashboard — web-based observability interface for monitoring agent activity

Architecture

Vigilance is built in Go with a modular bridge architecture:

unity_bridge/     Unity MCP bridge — captures viewport frames and game state
blender_bridge/   Blender MCP bridge — captures viewport frames and scene state
conductor/        Central orchestration — manages agent sessions and data flow
hud/              Heads-up display — web dashboard for real-time monitoring
frontend/         Web UI — MJPEG viewer with SVG overlay support
pkg/              Shared libraries — MCP protocol, streaming, utilities

Prerequisites

• Go 1.23+
• Unity or Blender (depending on which bridge you're using)
• An MCP-compatible AI agent (Claude Code, Codex, etc.)

Building

git clone https://github.com/B-A-M-N/Vigilance.git
cd Vigilance
go build -o vigilance ./cmd/vigilance

Or use the Makefile:

make

Running

./vigilance

The dashboard is available at http://127.0.0.1:8080.

Ecosystem Mode

To run the full Vibe ecosystem (Vigilance + BlenderVibeBridge + UnityVibeBridge):

./start_ecosystem.sh

Bridge Setup

Unity Bridge

• Place the Unity bridge script in your Unity project
• Configure the MCP endpoint to point to Vigilance
• The bridge captures viewport frames and forwards them via MJPEG

Blender Bridge

• Install the Blender bridge addon
• Configure the MCP endpoint to point to Vigilance
• The bridge captures 3D viewport frames and scene state

Key Features

• Local-first — all data stays on your machine, no cloud dependencies
• Real-time streaming — MJPEG with sub-second latency
• SVG overlays — dynamic bounding boxes and attention indicators overlaid on live video
• PII redaction — automatic scrubbing of secrets and sensitive data before display
• MCP protocol support — integrates with any MCP-compatible agent

Dependencies

Package	Purpose
mcp-go	MCP server implementation
gorilla/websocket	WebSocket connections for real-time streaming
google/uuid	Session and agent identification

Status

Vigilance is actively developed. Core frame ingestion and MJPEG streaming are operational. Intent tracking and cognitive DVR features are in development.