Claude Code Beyond Basics: Skills, MCP, and Open Models
Claude Code runs as a terminal agent, but most of its leverage comes from four extensions: running it inside an IDE, teaching it repeatable workflows with Skills, granting it new capabilities through MCP servers, and pairing it with compatible open-source models. This guide covers each.
Running Claude Code in an IDE
The CLI is fast, but an IDE like VS Code or Cursor adds a file explorer, inline editing, in-context selection, and a visual diff of proposed edits before you accept them.
- Install VS Code or Cursor.
- Open the Extensions view (
Ctrl+Shift+X), search for “Claude Code,” install the official Anthropic extension, and trust the publisher when prompted. - Launch it — the Claude icon in VS Code’s toolbar, or right-click → “Claude Code: Open” in Cursor. Sign in with your Claude subscription or Console account on first use.
A three-column layout — file explorer, active file, and the Claude Code chat panel — keeps everything in view without switching windows.
Skills: reusable workflows
A Skill is a set of custom instructions that tells Claude how to perform a specific task — a saved recipe, comparable to custom GPTs or Gemini Gems. Skills standardize repeatable work so you stop re-explaining the same requirements.
Each skill is a SKILL.md file in its own folder under .claude/skills/, optionally alongside examples, templates, or scripts.
The fastest way to start is to have Claude Code scaffold the structure and install the foundational Skill Creator skill:
Please set up a proper Claude folder structure for using Claude Skills, then install the Skill Creator skill: https://github.com/anthropics/skills/tree/main/skills/skill-creator
Claude Code creates .claude/skills/ and installs skill-creator inside it.
Invoking a skill:
– Manually — type / followed by the skill name (e.g. /skill-creator) to see its actions.
– Automatically — describe the task in plain language and Claude selects the relevant skill.
To build a new skill, just describe what you want — for example, “create a skill that turns any uploaded image into a three-haiku poem” — and skill-creator walks you through it.
MCP: borrowing external tools
The Model Context Protocol (MCP) is an open standard that lets an LLM talk to third-party tools and data sources, effectively borrowing their capabilities. Any service can publish an MCP server to expose its functionality to Claude Code.
Connect by asking Claude Code to do it:
Connect to the Hugging Face MCP server: https://huggingface.co/mcp?login
Manage servers from the terminal:
– claude mcp list — connected servers
– claude mcp get <name> — details for one server
– claude mcp remove <name> — disconnect
Once connected, Claude Code uses the server’s tools automatically when they’re relevant. After adding Hugging Face, a prompt like “What are the most popular 3D models on Hugging Face?” triggers its model_search tool without you naming it.
Skills and MCP together
Skills and MCP compound. A skill supplies the recipe — the workflow and desired output. An MCP server supplies the kitchen — the tools and data needed to run it. A “brand deck” skill, for instance, could call a design tool’s MCP server to generate slides and assemble them into a deck, automating an end-to-end creative task.
Pairing with open models
Claude Code also works with compatible open-source models such as SERA, which lets you keep coding workflows on self-hosted or lower-cost inference where that matters. See Open Coding Agents and SERA for the model side.
