Claude Code Series, Part 4: MCP (Model Context Protocol)

What is MCP?

MCP (Model Context Protocol) is a standardized way for Claude to access external services, databases, and tools. It’s a bridge between Claude Code and the outside world.

Think of it like this:

• Without MCP: Claude Code works within your local codebase
• With MCP: Claude Code can fetch docs from Notion, query GitHub issues, read web pages, and more — all while you’re in your terminal

MCP servers expose resources (data it can read) and tools (actions it can take).

Built-in MCP Servers

Claude Code ships with several pre-configured MCP servers:

1. Notion

Access your Notion databases and pages.

You: "What tasks are in our backlog?"

Claude: [Uses Notion MCP]
        [Lists all items from your Notion backlog database]

You: "Add a new task: 'Implement dark mode'"

Claude: [Creates a new Notion page in the backlog]

2. GitHub

Query issues, PRs, and repository data.

You: "Show me open bugs assigned to me"

Claude: [Fetches issues from GitHub API via MCP]
        [Lists bugs with labels and descriptions]

3. Web Fetch

Read and analyze web pages.

You: "What does the API documentation say about authentication?"

Claude: [Fetches the docs URL via MCP]
        [Reads and summarizes the authentication section]

4. Filesystem

Advanced file operations beyond Claude’s built-in tools.

You: "Find all TODO comments in the project"

Claude: [Uses MCP to recursively search the filesystem]
        [Lists all TODOs with line numbers]

Example: Using Notion MCP

Let’s say you have a Notion database of technical articles. Here’s how Claude Code uses it:

Step 1: Connect Notion Set up the Notion MCP server in your ~/.claude/mcp.json:

{
  "mcpServers": {
    "notion": {
      "command": "npx",
      "args": ["@modelcontextprotocol/server-notion"],
      "env": {
        "NOTION_API_KEY": "ntn_..."
      }
    }
  }
}

Step 2: Query Notion from Claude Code

You: "Which articles are about distributed systems?"

Claude: [Queries Notion database via MCP]
        [Filters by tag 'distributed-systems']
        Articles found:
        - Kafka Exactly-once Semantics
        - Raft Consensus Algorithm
        - Two-phase Commit Explained

Step 3: Act on the data

You: "Update our blog README with these articles"

Claude: [Reads the list from Notion]
        [Updates README with latest articles]
        [Commits and pushes the change]

Setting Up Custom MCP Servers

MCP servers are just programs that follow the MCP spec. You can run custom ones:

Example: Company Wiki MCP

Create a simple Node.js MCP server that reads your internal wiki:

// wiki-mcp-server.js
const { Server } = require('@modelcontextprotocol/sdk');

const server = new Server({
  name: 'wiki',
  version: '1.0.0',
});

// Expose wiki search as a tool
server.tool('search-wiki', {
  description: 'Search your internal wiki',
  inputSchema: {
    type: 'object',
    properties: {
      query: { type: 'string', description: 'Search term' }
    },
    required: ['query']
  },
  handler: async (params) => {
    // Query your wiki API
    const results = await fetch(`https://wiki.company.com/api/search?q=${params.query}`);
    return results.json();
  }
});

server.start();

Register it in ~/.claude/mcp.json:

{
  "mcpServers": {
    "wiki": {
      "command": "node",
      "args": ["/path/to/wiki-mcp-server.js"]
    }
  }
}

Now Claude Code can search your wiki:

You: "What's our policy on database backups?"

Claude: [Searches wiki via MCP]
        [Finds and reads the backup policy document]

Real-world MCP Workflow

Here’s how MCP transforms your workflow:

Before MCP:

You: "Add the features from our Notion board to the backlog"
Claude: I can't access Notion. You'll need to copy-paste them.

With MCP:

You: "Add the features from our Notion board to the backlog"

Claude: [Uses Notion MCP to fetch the board]
        [Reads all items]
        [Creates GitHub issues for each]
        [Posts a summary]
        ✓ Done!

MCP enables context flow: Notion → Claude Code → Your repo → GitHub

Common MCP Use Cases

Use Case	MCP Server	Benefit
Sync docs to code	Notion	Keep README and code in sync
Auto-create issues	GitHub	Turn Notion tasks into GitHub issues
Reference web docs	Web Fetch	Cite official docs while coding
Query databases	Custom	Access business logic and requirements
Search internal tools	Custom	Find examples in your codebase

Limitations & Best Practices

Limitations:

• MCP adds latency (network calls)
• Servers must be running and authenticated
• Complex operations need careful permission handling

Best practices:

• Keep MCP operations concise (don’t fetch 1000 rows)
• Use MCP to gather context, not as your only tool
• Combine with local file operations for speed
• Test MCP workflows before relying on them in CI/CD

Next: Building Reusable Skills

MCP connects you to external data. Skills let you create reusable, shareable tools that combine multiple tools and MCP servers into one command.

Next: Part 5 — Building Claude Skills