Skip to content
Dev MCP

Overview

Claude Code — The Development Gateway

Section titled “Claude Code — The Development Gateway”

All project development flows through Claude Code, Anthropic’s CLI-based agentic coding tool. It is not an IDE plugin or a chat window — it is a terminal-native co-pilot that reads your project files, executes commands, writes code and orchestrates external tools via MCP (Model Context Protocol) servers.

In our workflow, Claude Code acts as the single gateway to all project development. Engineers activate a platform agent, issue natural-language commands in the terminal, and Claude Code:

  • Reads design context directly from Figma via MCP
  • Generates code following the active agent’s platform conventions (Liquid, React/Remix, etc.)
  • Writes files to the correct locations per the agent’s directory structure
  • Opens a real browser to screenshot and visually verify output against Figma
  • Iterates on discrepancies until the implementation matches the design
  • Runs quality audits (performance, SEO, accessibility) on demand

Claude Code loads project context automatically at session start — including CLAUDE.md rules, the active platform agent’s conventions, remote Dev MCP universal rules, and the project’s CONVENTIONS.md — so every session begins with full platform awareness and project state.

What Is the Dev MCP Server?

Section titled “What Is the Dev MCP Server?”

The Dev MCP server is our company-hosted MCP server that provides Claude Code with five things:

  • Platform agents — define how to build for a specific stack. The Theme Developer carries Liquid/OS 2.0 conventions. The Hydrogen Developer carries React/Remix/Sanity conventions. Agents carry component patterns, build rules, validation checklists, and anti-patterns. (via activate_agent)
  • Universal conventions — mandatory rules, design token standards, MCP toolchain usage, and communication style that apply to all projects regardless of platform (via get_conventions)
  • Automated workflows — setup, build, design-system extraction, visual QA, performance/SEO/accessibility audits, maintenance. Commands are generic — they adapt to whichever agent is active. (via run_command)
  • Domain knowledge on demand — integration guides and workflow templates for platforms like SearchSpring, Nosto, and Tagalys (via list_skills / get_skill)
  • Project memory — context that persists across sessions per project, including agent preference, saved notes, command history, and project state (via save_project_note, get_project_context)

The Dev MCP server is one of six MCP servers initialised during project setup. Together, they give Claude Code the ability to read designs, write code, preview in a browser, manage git, query docs, and follow platform standards — all from a single terminal session.

What’s next

Section titled “What’s next”