MCP SERVERS - MODEL CONTEXT PROTOCOL

What is MCP?

Model Context Protocol (MCP) is a standard for connecting AI assistants to external tools and data sources. MCP servers expose capabilities that Claude Code can use through a standardized interface.

Tool Naming Convention:

MCP tools follow the pattern: mcp__<server>__<tool>
Examples:
- mcp__memory__create_entities
- mcp__rag-memory__ingest_url
- mcp__playwright__browser_click

Configuration:

MCP servers are configured in three locations:
1. Project config: .mcp.json (checked into repo for team)
2. Global config: ~/.claude/mcp.json (personal servers)
3. Settings: .claude/settings.json (managed policy)

Example .mcp.json:
{
  "mcpServers": {
    "rag-memory": {
      "command": "uv",
      "args": ["run", "rag-mcp-stdio"],
      "cwd": "/path/to/rag-memory"
    }
  }
}

Communication:

MCP servers communicate via:
- STDIO: Standard input/output (for local processes)
- SSE: Server-sent events (for HTTP servers)

Popular MCP Servers:

- Filesystem: File operations
- GitHub: Repository management
- Puppeteer: Browser automation
- Memory: Persistent knowledge graphs
- RAG Memory: Semantic search + knowledge graphs

Benefits:

- Standardized tool interface
- Reusable across projects
- Team sharing (via .mcp.json)
- Ecosystem of pre-built servers
- Language-agnostic (any language can implement)

Creating Your Own:

Use the MCP SDK in Python, TypeScript, or other languages to build custom servers that expose domain-specific tools to Claude Code.
