ARG MCP Server

The ARG way of letting AI agents like Claude, Cursor and Copilot use your application through a Model Context Protocol (MCP) server. The server reuses the features your app already has, so no business logic is duplicated.

Installation

npx skills add ARG-Software/arg-agent-skills --skill arg-mcp-server

What it does

  • The skill starts by mapping what agents need to do onto the use cases your app already has. Each MCP tool calls exactly one use case, with the same permissions and errors as your REST API.
  • The MCP server is a thin entry point next to your API, and architecture checks keep business logic and permission checks out of it. Your API and your agent interface grow together instead of drifting apart.
  • Remote servers are secured with OAuth, following the MCP authorization spec, and agents receive clear, typed errors instead of stack traces. Local servers run over stdio with no network exposure.
  • The server is tested with a real MCP client, and the skill writes setup instructions for Claude Code, Claude Desktop, Cursor, VS Code, Codex and OpenCode.

Used when

The skill applies when adding MCP support to an API or back end, making an app usable from AI assistants, exposing use cases to agents, choosing between stdio and HTTP transport, adding OAuth to a remote MCP server, or testing and documenting an MCP server.

This skill is how ARG makes existing applications usable by AI agents. The worked example comes from ARG's NestJS MCP starter, where a REST API, an MCP server and a background worker share one application layer, and the approach applies to any Node.js or TypeScript back end.