Cellar FAQs

Question 1

What is the Model Context Protocol (MCP) and how does Cellar integrate with it?

Accepted Answer

The Model Context Protocol (MCP) is a standard for AI agents to interact with their environment. Cellar ships as a first-class MCP server, providing core tools like `cel_see`, `cel_act`, `cel_think`, and `cel_perceive`, making it readily compatible with MCP-native clients such as Claude Code and Cursor.

Question 2

Can Cellar automate tasks beyond web browsers?

Accepted Answer

Yes, a key differentiator of Cellar is its ability to extend automation beyond web browsers to native macOS applications, terminals, and spreadsheets. It operates anywhere an accessibility tree is available, allowing agents to seamlessly interact with the entire desktop environment using a single runtime.

Question 3

Is Cellar compatible with various AI models and privacy-sensitive workflows?

Accepted Answer

Absolutely. Cellar is model-agnostic, supporting various LLM providers, including local models like Gemma 4. It's built for privacy-sensitive teams, running offline with no cloud or telemetry, ensuring your keys and data remain on your machine.

Question 4

What is Cellar and its primary purpose?

Accepted Answer

Cellar is a robust computer use runtime for AI agents specifically designed for macOS. Its primary purpose is to enable AI agents with structured perception, reliable actions, and continuous awareness across both desktop applications and web browsers, solving the limitations of vision-only approaches.

Question 5

How does Cellar achieve superior perception for AI agents?

Accepted Answer

Cellar utilizes a unique hybrid perception approach, fusing data from accessibility trees, Chrome DevTools Protocol (CDP), native macOS APIs, and falling back to vision only when structured data is unavailable. This 'structure first' method provides more accurate, reliable, and cost-effective context than screenshot-only agents.

Cellar

Cellar

Key Features

Use Cases

Key Features

Use Cases