MCP: Extending Claude's Reach

Out of the box, Claude Code can read files, run shell commands, and search your codebase. That's powerful, but it's a closed system. What if Claude could query your production database directly? Or control a browser? Or interact with your GitHub issues without you copy-pasting context?

That's what MCP does.

What is MCP?

The Model Context Protocol is an open standard for connecting AI models to external tools and data sources. Think of it as USB for AI -- a universal interface that lets Claude plug into anything.

An MCP server exposes tools (functions Claude can call), resources (data Claude can read), and prompts (templates Claude can use). Claude Code acts as an MCP client, discovering and calling these tools as needed.

Adding MCP Servers

There are two ways to add an MCP server: the CLI command, or editing a config file directly.

Via CLI

# stdio transport (local process)
claude mcp add --transport stdio my-server npx -y @modelcontextprotocol/server-filesystem /path/to/dir
 
# HTTP transport (remote server)
claude mcp add --transport http my-api-server https://api.example.com/mcp
 
# SSE transport (server-sent events)
claude mcp add --transport sse my-sse-server https://example.com/mcp/sse

Transport Types

Most MCP servers you'll use are stdio -- they run as local processes on your machine.

Configuration Files

MCP servers are configured in two places:

Project-level: .mcp.json

Lives in your project root. Committed to git so your whole team gets the same tools.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "./src"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "your-token-here"
      }
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "DATABASE_URL": "postgresql://localhost:5432/mydb"
      }
    }
  }
}

User-level: ~/.claude.json

Personal config that applies to all your projects. Good for tools you always want available.

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp", "--browser", "chromium"]
    }
  }
}

Lazy Loading and Tool Search

When you have many MCP servers configured, loading all their tools at startup wastes context window space. Claude Code solves this with lazy loading via the ENABLE_TOOL_SEARCH environment variable (enabled by default).

With tool search enabled, Claude doesn't load all MCP tools upfront. Instead, it discovers tools on-demand -- when a task requires a specific capability, Claude searches for matching tools and loads only what it needs.

This means you can have dozens of MCP servers configured without paying a context cost for tools you're not using in the current conversation.

Practical Examples

Filesystem MCP

Gives Claude structured access to specific directories with fine-grained permissions:

claude mcp add --transport stdio filesystem npx -y @modelcontextprotocol/server-filesystem /Users/me/projects

GitHub MCP

Lets Claude interact with GitHub issues, PRs, and repositories directly:

claude mcp add --transport stdio github npx -y @modelcontextprotocol/server-github

Now Claude can list issues, create PRs, read PR comments, and search code -- all without you leaving the terminal.

Database MCP

Connect Claude to a Postgres database for direct queries:

claude mcp add --transport stdio postgres npx -y @modelcontextprotocol/server-postgres

Claude can then inspect schemas, run queries, and help debug data issues with full context.

Playwright MCP

Give Claude a browser it can control for testing and verification:

claude mcp add --transport stdio playwright npx -y @playwright/mcp --browser chromium

Claude can navigate pages, take screenshots, fill forms, and verify your UI works correctly.

Exercises

# Add the server
claude mcp add --transport stdio filesystem npx -y @modelcontextprotocol/server-filesystem ./src
 
# Verify it's configured
cat .mcp.json
 
# Start Claude Code and ask:
# "What files can you see via the filesystem MCP?"

# Add with environment variable
claude mcp add --transport stdio github npx -y @modelcontextprotocol/server-github
 
# Then edit .mcp.json to add the env block:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token_here"
      }
    }
  }
}

# Start Claude Code and ask:
# "List open issues on my-org/my-repo"

What's Next

MCP extends Claude's capabilities externally. In the next section, we'll look at how sub-agents extend Claude's capabilities internally -- by running multiple Claude instances in parallel.