~/agent-config
Learn AI
AgentsJul 15, 20261 min read

From Chatbot to Agent: How AI Coding Tools Take Action

A chatbot answers. An agent does. Here's the loop that turns a language model into a tool that can read files, run commands, and edit your codebase — and how to set one up.

A plain chatbot can only produce text. An agent wraps that same model in a loop and gives it tools — so instead of describing a fix, it can open the file, make the change, run the tests, and read the results.

The agent loop

  • Observe — the model reads the task and the current state.
  • Decide — it chooses a tool to call, or answers directly.
  • Act — the tool runs (read a file, run a command, call an API).
  • Repeat — the result feeds back in, and the loop continues until done.

Where tools come from

Tools are exposed to the model through a standard interface. The Model Context Protocol (MCP) has become a common way to plug capabilities — GitHub, a filesystem, a database — into any agent that speaks it. Add an MCP server and every compatible agent can suddenly use it.

A sane starting setup

You don't need every tool at once. A filesystem server for local files, a GitHub server for repo context, and a couple of good rules for code review and testing will take you a long way. Browse the catalog and copy what you need.

Get started with these

Drop-in configs from the catalog to put this into practice.

Skill

Code Review

Reviews diffs for correctness, security, and style before you open a PR. Flags risky changes, missing tests, and edge cases the author may have missed…

#review #quality #pr +1

Cursor, Claude Code, Windsurf +3

MCP

Filesystem MCP

Official MCP server for controlled local file access. Set allowed directories in the args — only paths listed here can be read or written by the agent…

#mcp #filesystem #local +1

Cursor, Claude Code, Windsurf +2

Details
MCP

GitHub MCP

Lets your agent read issues, PRs, and repo metadata from GitHub. Add a personal access token with repo scope to .cursor/mcp.json (or your agent's MCP…

#mcp #github #git +2

Cursor, Claude Code, Windsurf +2

Details
Skill

Test-Driven Development

Drives implementation from failing tests first. Writes the test, watches it fail, makes it pass, then refactors — keeping coverage honest as you build…

#testing #tdd #quality +1

Cursor, Claude Code, Windsurf +3