create-mcp-servers_skill

This skill guides you to create production-ready MCP servers exposing tools, resources, and prompts for Claude via secure, scalable integrations.

Shell

1.2k

GitHub Stars

1

Bundled Files

2 months ago

Catalog Refreshed

4 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 veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill glittercowboy/taches-cc-resources --skill create-mcp-servers

SKILL.md5.7 KB

Overview

This skill helps you create production-ready Model Context Protocol (MCP) servers that expose tools, resources, and prompts to Claude. It supports both TypeScript and Python flavors and includes patterns for small and large APIs, security guidance, and exact terminal commands. Use it to integrate custom APIs, data sources, or authentication flows so Claude can call them reliably.

How this skill works

The skill guides you through a validated workflow: project setup, operation design, secure configuration, and verification in Claude Code/Desktop. It enforces architecture choices based on operation count (traditional vs on-demand discovery) and provides templates, scripts, and commands to initialize Python or TypeScript servers. It also includes checks for environment variables, absolute paths, and logging to ensure robust, deployable servers.

When to use it

Building a new MCP server to expose API endpoints, file data, or prompt templates to Claude.
Extending an existing server with new operations, resources, or OAuth flows.
Integrating third-party services so Claude can call them as tools.
Preparing a server for production with secure env vars, logging, and validation.
Troubleshooting connection, transport, or environment issues for an MCP server.

Best practices

Never hardcode secrets; use environment variables and ${VAR} expansion in configs.
Run projects in isolated directories (one server per ~/Developer/mcp/{server-name}/).
Use absolute executable paths found via which (e.g., which uv, which node, which python).
Prefer uv for Python projects to manage venvs; avoid passing cwd via args—use cwd property.
Choose architecture by operation count: 1–2 operations = traditional; 3+ = on-demand discovery.

Example use cases

Create a Python MCP server that wraps a private API and exposes search and fetch tools to Claude.
Build a TypeScript server that streams large dataset responses and implements pagination/truncation.
Add OAuth stdio isolation for third-party auth flows used by an MCP server.
Migrate a flat toolset into on-demand discovery with operations.json and meta-tools.
Run automated validation checks and logging before registering the server in Claude Desktop.

FAQ

Use traditional for 1–2 discrete operations. Use on-demand discovery with meta-tools plus operations.json when exposing 3 or more operations or when operations are dynamic.

What if my server command shows 'command not found'?

Use absolute paths returned by which (e.g., which uv || which node) and update your mcp config to reference that path.

Can I store API keys in config files?

No. Always use environment variables and ${VAR} expansion; never commit keys or paste them into chat.