dimpagk92/cellar

What it does

Cellar is a computeruse runtime that fuses multiple input streams—accessibility tree data, Chrome DevTools Protocol, and computer vision—into unified structured context for AI agents. It provides four core MCP tools (see/act/think/perceive) that enable sophisticated interaction with digital interfaces.

Key Features

• Hybrid input fusion combining accessibility trees, CDP, and vision
• Per-element confidence scoring for reliable interaction
• Continuous awareness engine called Cortex
• Freshness detection and side-effect monitoring
• Offline-capable with Ollama support
• Four MCP tools: see, act, think, perceive

How to set up

Cellar operates as an MCP (Model Context Protocol) server. Installation requires setting up the server from the GitHub repository (dimpagk92/cellar) and configuring it with your preferred AI client. It supports offline operation with Ollama and can be integrated with compatible AI platforms that support MCP servers.