mcp-integration_skill

This skill guides you to integrate Model Context Protocol servers into Claude Code plugins, enabling external tool access and secure, scalable configurations.

Shell
Official

73.8k

GitHub Stars

1

Bundled Files

3 weeks ago

Catalog Refreshed

2 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstart where the catalogue uses aiagentskills.

npx veilstart add skill anthropics/claude-code --skill mcp-integration

SKILL.md12.2 KB

Overview

This skill guides developers through integrating Model Context Protocol (MCP) servers into Claude Code plugins to expose external tools and services. It focuses on configuration patterns, server types (stdio, SSE, HTTP, WebSocket), authentication, lifecycle, and testing. Use it to reliably add external APIs, local servers, or real-time services as tools inside Claude Code.

How this skill works

The skill explains where to declare MCP servers (recommended .mcp.json or plugin.json inline), how Claude Code starts or connects to servers, and how tools are discovered and named. It covers transport differences: stdio spawns local processes, SSE uses server-sent events with OAuth, HTTP uses REST with tokens, and ws uses WebSocket for real-time. It also explains environment variable expansion, tool naming, and lifecycle management.

When to use it

Adding an MCP server to a Claude Code plugin ("add MCP server" / "integrate MCP").
Exposing external APIs, databases, or file systems as tools inside Claude Code.
Configuring authentication flows (OAuth or token-based) for an external service.
Bundling a local MCP process with a plugin using stdio.
Connecting real-time services via WebSocket or hosted services via SSE.

Best practices

Prefer .mcp.json at plugin root for multiple servers and clearer maintenance.
Use ${CLAUDE_PLUGIN_ROOT} for portable paths and document required env vars in the README.
Use HTTPS/WSS and avoid hardcoding tokens; load secrets from environment variables.
Pre-allow specific MCP tools in command frontmatter instead of using wildcards.
Test connectivity with /mcp and use claude --debug for startup and auth logs.

Example use cases

Bundle a local stdio MCP server to provide filesystem and local DB access for commands.
Connect to an SSE-hosted service (Asana, GitHub) and let Claude Code handle OAuth on first use.
Use an HTTP MCP endpoint with Bearer token auth to call a custom REST backend.
Attach a WebSocket MCP server for low-latency streaming or push notifications.
Create an autonomous agent that queries a database via MCP tools, processes results, and generates reports.

FAQ

Use .mcp.json for multiple or complex servers; use plugin.json inline for a simple single-server plugin.

How are MCP tool names formed and how do I allow them?

Tools are auto-prefixed: mcp__plugin_<plugin-name>_<server-name>__<tool-name>. Pre-allow specific tool names in command frontmatter for security rather than wildcards.