AI agent tools: GitHub MCP, Browser Use and memory

This tutorial connects seven tools to the Codex app through MCP servers, CLI skills and plugins. The presenter argues that an AI agent benefits from carefully chosen connections and warns that too many tools can confuse tool selection.

GitHub MCP handles repository tasks, with a demonstration that retrieves open issues and pull requests. Composio groups account connections behind one MCP server and discovers tools when needed. Context7 supplies current framework documentation for a coding assistant; the example requests recent React documentation.

The browser and search sections explain different jobs. Browser Use can delegate a task to a cloud agent, provide a remote browser or control a local browser through its CLI. The local example requires remote debugging and reuses the browser's signed-in session. The presenter says this route is slower and unsuitable for a headless environment. Browser Use sponsors the video.

Exa searches for information, while Firecrawl extracts page content. A combined example researches local AI models and turns the retrieved material into a short summary. These are connected-service demonstrations, not instructions to run models locally.

Mem0 adds memory across sessions and agents. The setup requires an account, and the demonstration saves a dark-mode preference. The tutorial shows installation and authorization steps, but does not establish an offline workflow.