Engineering an MCP (Model Context Protocol) Tool for Cursor AI: Step-by-Step Guide

Introduction : MCP (Model Context Protocol) provides a standardized way for applications to supply context to LLMs, separating context provision from LLM interaction.

Problem background : When using Cursor AI to call MCP services, developers encounter two main pain points: (1) official SDK examples cannot be directly adapted to Cursor AI, and (2) repetitive MCP code development (protocol parsing, error handling, logging) increases maintenance cost and leads to inconsistent code styles.

Problem analysis : The lack of a unified engineering solution causes duplicated boilerplate across MCP services and hampers team collaboration.

Solution : Engineer an MCP tool with a systematic workflow: (1) encapsulate the MCP SDK into a reusable package, (2) define a configuration file, (3) implement the core PSD slicing logic, (4) describe input schemas with JSON Schema, (5) provide helper utilities, (6) create the main service entry, (7) add build scripts, (8) integrate the tool into Cursor AI as a command‑type server.

Implementation process : Steps 1‑6 are illustrated with screenshots of configuration files (src/config.ts), tool logic (src/tools/psd-slice.ts), schema definitions (src/schemas/input.ts), helper functions (src/utils/helpers.ts), and the service entry (src/index.ts). Build script configuration and execution (npm run build) produce a compiled script in the build directory.

Deployment in Cursor AI : Add a new MCP server entry, set its type to command , point to the built script (e.g., /Users/KevinKwok/mcp/psd-slice-mcp/build/index.js), enable YOLO mode, and the tool becomes callable from an Agent conversation. The result shows layer hierarchy matching the original PSD, with previews.

Common pitfalls : (1) Server must be instantiated with required parameters (otherwise “Server does not support tools” error). (2) File‑path parameters need absolute‑path descriptions to avoid missing files. (3) Ensure all dependencies are installed for the built artifact.

Summary : The guide demonstrates a complete end‑to‑end MCP tool development, from problem definition to code generation, build, and integration with Cursor AI.

Outlook : Future work aims to build an intelligent MCP development framework that automatically generates config, tools, and schemas from natural‑language descriptions, enabling seamless usage in MCP clients.