What is MCP?
How Claude Code connects to external tools, databases, and APIs
⏱ Est. ~5 min
01 · Read
Claude Code can read your files, write code, and run commands — but only on your local machine. What if you want Claude to see your GitHub issues? Query your database? Look at Sentry error logs? Read your team's Slack messages?
Right now Claude can't. It's stuck in your terminal, cut off from the tools you use every day.
MCP changes that.
💡 Picture thisThink of Claude Code today as a super-skilled contractor who can only work in one room. MCP is like giving them keys to the whole building — the database room, the monitoring room, the project management office. Same contractor, but now they can walk between rooms and get things done across your entire workflow.
Key points
- Claude Code alone only accesses your local files and terminal
- MCP connects Claude to external tools, databases, and APIs
- MCP is the Model Context Protocol — an open standard
- It works with Claude Code, Claude Desktop, and other AI tools
02 · Read
MCP stands for Model Context Protocol. It's an open standard — think of it as USB-C for AI applications.
Just like USB-C gives you one universal port for phones, monitors, hard drives, and keyboards, MCP gives AI tools one universal protocol to connect to GitHub, databases, Sentry, Slack, and hundreds of other services.
Without MCP, every AI tool would need to build custom integrations for every service. Expensive and brittle. MCP solves that: build one MCP server for your service, and every MCP-compatible AI tool can use it.
Key points
- MCP = Model Context Protocol, Anthropic's open standard
- "USB-C for AI" — one protocol to connect to many services
- MCP servers expose tools that Claude Code can call
- Hundreds of MCP servers already exist (GitHub, Postgres, Sentry, Notion, etc.)
03 · Code example
Once an MCP server is connected, you can ask Claude Code to do things like this — all in natural language.
Example prompts using MCP
"Add the feature described in JIRA issue ENG-4521 and create a PR on GitHub."
"Check Sentry for the most common errors in the last 24 hours."
"Find emails of 10 users who used feature X, based on our PostgreSQL database."
"Review PR #456 on GitHub and suggest improvements."
"Create Gmail drafts inviting these users to a feedback session."
Each one needs Claude to talk to an external service (JIRA, Sentry, PostgreSQL, GitHub, Gmail). Without MCP, Claude can't do any of them. With the right MCP server connected, it can do all of them.
04 · Read
MCP servers connect to Claude Code in one of three ways — called a transport.
HTTP is the recommended option for cloud services. The server runs remotely, and Claude connects over the network. The most popular MCP servers (GitHub, Sentry, Notion) use HTTP.
Stdio runs a local process on your machine. Good for tools that need direct system access, like database connectors or custom scripts. The command runs as a subprocess — Claude communicates through standard input/output.
SSE (Server-Sent Events) is an older transport that's now deprecated. You might see it in old tutorials, but new setups use HTTP.
Key points
- HTTP: remote server, recommended for cloud services (GitHub, Sentry, Notion)
- Stdio: local process on your machine (databases, custom scripts)
- SSE: deprecated, use HTTP instead
- Most MCP servers you'll use are HTTP — just a URL
05 · Quiz
Think about what Claude can access today vs. what it can access with MCP. What does MCP make possible that Claude Code can't do on its own?
- Read files in your project
- Run terminal commands
- Write code to files
- Query an external database or look at GitHub issues
06 · Match
You've seen the three transports. Match each one to when you'd use it.
(This section is interactive — please enable JavaScript.)
Other lessons in this chapter
⚠ The full interactive experience needs JavaScript. Please enable it and reload.
※ This is an independent Traditional Chinese teaching project — not an official Anthropic product. Claude™ is a trademark of Anthropic, PBC.