Why aren't my MCP tools visible in Claude Code?

This is usually caused by failing to restart the Claude Code session after running the 'claude mcp add' command, or using the legacy manual config file instead of the CLI tool.

What is the recommended MCP isolation pattern?

The Daemon (SSE) pattern is the default recommendation as it provides the best balance of fast startup (1-2s), zero token overhead, and persistent state sharing.

How much token space can these patterns save?

For heavy MCP servers like Serena or Playwright, these patterns can save between 6,000 to 10,000 tokens per session by offloading tool definitions from the primary context window.

When should I use the Subprocess strategy?

Subprocess isolation is best for offline or air-gapped environments where network-based daemons aren't allowed, or for tools used very infrequently (less than twice per session).

MCP Gateway Patterns

Name: MCP Gateway Patterns
Author: chkim-su

bychkim-su

Developer Tools

Implements optimized architectural patterns for MCP integrations to reduce token overhead and improve performance.

About

MCP Gateway Patterns provides a comprehensive framework for managing Model Context Protocol (MCP) servers within Claude Code. By implementing specialized isolation strategies—Daemon (SSE), Agent Gateway, and Subprocess—this skill helps developers minimize context window bloat, potentially saving over 10,000 tokens for complex toolsets like Playwright or Serena. It includes empirical decision guides to help you choose the right pattern based on startup latency, state-sharing requirements, and environmental constraints such as air-gapped systems or serverless deployments.

Key Features

Standardized 2-layer protocol for intent-based tool execution (QUERY, ANALYZE, MODIFY, EXECUTE)
0 GitHub stars
Triple-strategy isolation: Daemon (SSE), Agent Gateway, and Subprocess
Token usage optimization saving up to 100% of tool-definition overhead
Automated MCP registration guidance for user and project scopes
Empirical performance benchmarks for latency and state sharing

Use Cases

Deploying MCP integrations in restricted serverless or air-gapped environments
Optimizing heavy MCP servers with high tool counts to prevent context window saturation
Standardizing tool naming conventions and access rules across complex Claude Code projects

About

Key Features

Standardized 2-layer protocol for intent-based tool execution (QUERY, ANALYZE, MODIFY, EXECUTE)
0 GitHub stars
Triple-strategy isolation: Daemon (SSE), Agent Gateway, and Subprocess
Token usage optimization saving up to 100% of tool-definition overhead
Automated MCP registration guidance for user and project scopes
Empirical performance benchmarks for latency and state sharing

Use Cases

Deploying MCP integrations in restricted serverless or air-gapped environments
Optimizing heavy MCP servers with high tool counts to prevent context window saturation
Standardizing tool naming conventions and access rules across complex Claude Code projects