Editorial update (April 2026): My position on MCP has evolved since I wrote this. After three more months of daily use, I removed the GitHub MCP server and replaced it with gh CLI — the context cost wasn’t worth the marginal integration benefit. The argument is in CLI > MCP. Supabase, Playwright, Context7, and Figma MCP remain part of my daily workflow. Read this post as the “why start with MCP” case and the CLI piece as “where I ended up.”

What Is MCP?

The Model Context Protocol (MCP) lets AI assistants interact with external tools and services through a standardized interface. Instead of copying data between tools manually, MCP servers expose capabilities that Claude can call directly — an interface layer between Claude and your development ecosystem.

GitHub MCP Server

The most immediately useful MCP server for developers. It gives Claude direct access to:

  • Create and manage pull requests
  • Read and comment on issues
  • Search code across repositories
  • View PR review status and checks

Instead of copying PR URLs and describing changes, you can say “review the open PR on this repo” and Claude reads it directly.

Playwright MCP Server

Browser automation for verification. After building a feature, Claude can:

  • Navigate to your dev server
  • Take screenshots
  • Check for console errors
  • Verify UI elements render correctly

This closes the feedback loop — build, verify, iterate — without leaving the terminal.

Supabase MCP Server

Direct database interaction for projects using Supabase:

  • Run SQL queries
  • Apply migrations
  • Generate TypeScript types
  • Check advisor recommendations

Particularly useful during development when you’re iterating on schema changes.

Context7 Documentation Server

Fetches current library documentation on demand. Even for libraries Claude knows well, the docs may have changed since training. Context7 ensures Claude works with the latest API surface.

Setting Up MCP Servers

MCP servers are configured in .claude/settings.json:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "your-token"
      }
    }
  }
}

Conclusion

MCP servers extend Claude’s reach beyond the filesystem. Start with GitHub and Playwright — they cover the most common workflows. Add others as your needs grow.