Agentic AI Learning Portal

Appearance

MCP for Builders ​

What you'll learn

  • MCP turns external systems (databases, Notion, GitHub) into tools your AI coding agent can call directly
  • Business data MCP servers connect Claude Code to real databases and tools your team already uses
  • Multi-tool workflows combine servers in one session for tasks that span code, data, and documentation
  • Production MCP configs separate dev from CI, protect credentials, and keep projects consistent across teams

You know Claude Code can write code. But most real work happens outside your .ts and .py files. Your customer list lives in Postgres. Your product specs are in Notion. Your deployment history is in GitHub issues. Claude Code can't reach any of it unless you connect those systems.

That is what MCP (Model Context Protocol) does. One claude mcp add command, and your database becomes a tool Claude can query. Your Notion workspace becomes something Claude can read and update. Your GitHub repos become context Claude pulls in before it writes a line of code.

This is not the "what is MCP" course. If you have never connected an MCP server before, start with MCP Servers in the Claude Code course, which covers basic installation, scopes, and the three-line version of connecting a server. This course assumes you have done that once or twice and are ready to use MCP for actual business work: querying production databases, combining multiple servers in one workflow, and managing MCP across a team.

By the end, you will have a working setup that connects Claude Code to your own data, composites multiple servers, and runs safely in CI.

Prerequisites ​

You need three things to follow this course:

  • Claude Code installed and authenticated. You have run claude in a terminal and completed at least one interactive session. The Claude Code course covers installation and first use.
  • A database or business tool to connect. Lesson 1 uses Postgres and Notion. You can follow along with either. If you have neither, a local Postgres instance (via Docker or apt install postgresql) takes five minutes to set up, and you can use the same database through all three lessons.
  • Basic comfort with environment variables and shell config files. You will edit .mcp.json files, export env vars, and configure ~/.claude.json. Nothing beyond what a working developer does every day.

What you will not need: An Anthropic API key (Claude Code handles auth through your Anthropic account), a Notion admin account (any Notion account with API access works), or prior MCP experience beyond the basic connection steps.

Course structure ​

Connecting Business Data to Claude Code is the practical alternative to the basic MCP lesson. It covers two MCP servers that connect to real business systems: Postgres (for relational data) and Notion (for documents and wikis). You will configure both, query live data, and learn which problems each solves. Along the way, you will pick up connection string hygiene, read-only vs. read-write scoping, and debugging dead servers.

Multi-Tool MCP Workflows shows you how to run multiple MCP servers in a single Claude Code session and get them to cooperate. The lesson walks through a real PR review workflow where GitHub MCP retrieves the diff, Filesystem MCP reads the affected project files, and Claude synthesizes a review. Then a documentation update workflow where Claude reads the source tree through Filesystem, checks the issue tracker through GitHub, and writes updated docs. You will learn how to avoid context flooding, manage server overlap, and decide which tools should handle which parts of a task.

MCP in Production moves past the single-machine dev setup. It covers .mcp.json for project-scoped configuration that your team can commit, CI/CD integration with --mcp-config and --strict-mcp-config, and security boundaries: which server gets which credentials, how environment variables protect secrets, and why you should never put a read-write database connection in a shared config. The lesson ends with a CI job that connects to Postgres, runs a validation query, and gates a deployment.

Final build ​

By the end of this course you will have a multi-server MCP configuration that Claude Code loads in any project. It talks to your database for data questions, your Notion workspace for documentation, your GitHub repos for PR context, and your filesystem for the code itself. The configuration is split across personal and project scopes, credentials are in environment variables (never committed), and a CI job in .github/workflows/ runs the same setup to validate data before deployment. That is a real bridge between your AI coding agent and your company's actual data surface. No demos, no mock data.