Deploy MCP
An open-source Model Context Protocol (MCP) server enabling AI editors (Cursor, Claude Desktop, Antigravity, VS Code) to test, build, and deploy full-stack web applications to Vercel in seconds with zero server overhead and zero secret leakage.
How I got this idea
A friend of mine who had just started learning web development called me urgently asking for help. He had built a project with a fast-approaching deadline and had very little time to get it live. I wanted to help him deploy, but trying to explain the entire deployment workflow to someone completely new over a quick call — installing CLIs, handling Vercel logins, configuring Git remotes, setting environment keys, and fixing cryptic build errors — was overwhelming and took too much time. Seeing how stressed he was, it hit me: why does getting a website live still have so much friction, especially when beginners and pros alike are already building with AI editors? If someone is coding in Cursor, Claude Desktop, or Antigravity, the AI editor should just be able to test the build, manage secrets, and deploy to Vercel autonomously in seconds with a single prompt. That exact scramble to help my friend was the spark that led me to build Deploy MCP.
Problem
- AI coding assistants can generate full web applications in seconds, but taking them live requires context switching to external terminals, manual Vercel CLI commands, browser auth screens, and manual environment variable configuration.
- Exposing cloud deployment credentials or .env secrets to LLM prompt context poses severe security risks and credential leak vulnerabilities.
- Pushing untested code directly to cloud hosting exhausts build quotas on avoidable syntax, missing dependencies, or framework configuration errors without automated diagnosis.
Solution
- Engineering Approach: Built an autonomous, zero-trust deployment runtime conforming to Anthropic's Model Context Protocol (MCP) using Stdio JSON-RPC transport.
- Architecture: Developed a Node.js/TypeScript daemon exposing 20 granular tools (including smart_deploy, detect_project, diagnose_build_failure, sync_env), featuring an AST project detector, a sub-2-second local pre-flight build sandbox, and a local encrypted TLS gateway to the Vercel REST API v13.
- What I Personally Built: Programmed the full MCP server specification in TypeScript, implemented AST framework auto-detection for Next.js/Vite/Remix/Astro, built the local secret isolation barrier that masks environment values from LLM prompts, designed the self-healing error diagnostic parser for cloud build logs, and published the official deploymcp npm package and landing portal.
- Current Status & Result: Published production npm package (v2.0.0) with global CLI support (npx deploymcp setup), active open-source GitHub repository, and live web portal running at deploymcp.vercel.app with 20 autonomous deployment tools.
Autonomous MCP Pipeline & Zero-Trust Architecture
npx -y deploymcp setupHardest Technical Challenge
Achieving zero-trust secret isolation while enabling autonomous cloud environment synchronization without leaking sensitive credentials into the LLM context window. Solved by architecting a local cryptographic enclave middleware that intercepts environment configurations before they reach the Model Context Protocol layer. The AST scanner extracts and presents only sanitised variable names (e.g., DATABASE_URL, JWT_SECRET) to the AI assistant for parity validation, while the underlying values are encrypted and transmitted directly via authenticated local TLS pipelines to Vercel's REST API endpoints, completely isolating sensitive tokens from prompt memory.
Project Demo / Interface
What I learned
- Anthropic's Model Context Protocol (MCP) specification and Stdio JSON-RPC bidirectional communication.
- Zero-trust credential shielding, local AST static analysis, and secure token isolation architectures.
- Automated build telemetry parsing and algorithmic self-healing diagnostics for CI/CD pipelines.
- npm package distribution, CLI tool ergonomics (npx deploymcp), and edge hosting on Vercel.