Your First Agent: Beyond Chat

Tutorial: setting up Claude Code, MCP servers, and your first persistent agent.

Chat vs. Agent

Most people use AI the same way they use Google — with better grammar. Ask a question. Get an answer. Copy-paste it somewhere. Start a new conversation. Repeat.

That's chat. It's useful, but it throws away everything between sessions. Every conversation starts from zero. The AI doesn't know your project, your preferences, your tools, or your history.

An agent is different. An agent has:

- Persistent context: It reads your files, knows your project structure, and maintains understanding across sessions - Tool access: It can run commands, call APIs, read databases, and interact with external services - Memory: It remembers your preferences, your working patterns, and the decisions you've made - Autonomy: It can take multi-step actions, not just answer questions

The shift from chat to agent is the shift from "answer my question" to "understand my work and help me do it better."

Setting Up Claude Code

Step 1: Install. Claude Code runs in your terminal. It reads your filesystem directly — no copy-pasting code snippets into a browser.

```bash npm install -g @anthropic-ai/claude-code ```

Step 2: Point it at your project. Navigate to any project directory and run `claude`. It scans the directory, understands the file structure, and starts with context about your codebase.

Step 3: Create a CLAUDE.md. This is the file that turns Claude from a generic assistant into YOUR assistant. It contains:

- Your preferences (how you like code formatted, what frameworks you use) - Project context (what this repo does, how it's organized) - Instructions (what to always/never do) - Memory pointers (where to find project history, decisions, people)

A good CLAUDE.md is the difference between "helpful AI" and "AI that works like a team member who's been on the project for months."

Step 4: Add MCP servers. MCP (Model Context Protocol) connects your agent to external tools. Canvas LMS, Slack, GitHub, knowledge graphs, databases — any tool with an MCP server becomes a capability your agent can use.

What Changes

Once you have an agent instead of a chat window, the workflow shifts fundamentally.

You don't copy-paste code and ask "what's wrong?" You say "the tests are failing" and the agent reads the test output, finds the failing tests, reads the relevant code, identifies the issue, and proposes a fix — all in one step.

You don't describe your project every time. The agent reads CLAUDE.md and knows the architecture, the conventions, and the current priorities.

You don't manually coordinate between tools. The agent calls the GitHub API to check PR status, reads the Slack thread about the deployment, and updates your task tracker — because MCP servers give it access to all three.

Thirty minutes to set up. Changes everything about how you work with AI. The gap between "person who uses AI" and "person who directs AI agents" is the gap that will define knowledge work careers for the next decade.