Claude Code's MCP integration is one of the most practical examples of MCP in production. The patterns there generalise.
The integration
Claude Code reads ~/.claude.json (and project-level configs) for MCP servers. Each server registered becomes available to the assistant. Tools the server exposes are tools the assistant can call.
Common servers
- Filesystem. File operations within configured paths.
- GitHub. PR creation, issue management.
- Internal APIs. Team-specific tooling.
- Web fetch. Reading external pages.
- Database. Read-only queries against analytics.
Each is an MCP server. Each follows the conventions covered in companion articles on MCP design.
A real workflow
A team's developer flow:
- Local Claude Code session.
- Filesystem MCP for code editing.
- GitHub MCP for PR creation.
- Internal-tools MCP for ticket integration.
The assistant orchestrates across these. The team's productivity scales with the MCP investment.
Patterns that work
- Servers that respect the user's permissions.
- Servers that surface what they did clearly.
- Servers that fail gracefully.
- Servers that compose well in agent plans.
Trade-offs
- More servers = more capability, more complexity.
- The right portfolio depends on the team's workflow.
Most teams have 5-10 servers in active use. More than that becomes hard to reason about.
What we won't ship
MCP servers in Claude Code that don't match the conventions.
Servers without proper auth and audit.
Configurations that aren't versioned per project.
Close
MCP and Claude Code is the agentic-engineering pattern in practice. The assistant orchestrates; the MCP servers do the work. The conventions covered in companion MCP articles apply directly.
Related reading
- MCP in 10 minutes — primer.
- Plan vs. act: the agent loop — surrounding pattern.
- A senior engineer's day with Claude Code — companion context.
We build AI-enabled software and help businesses put AI to work. If you're integrating MCP with Claude Code, we'd love to hear about it. Get in touch.