How I Built an Agentic SDLC Demo with Cursor, MCP Servers, Playwright, GitHub, and Vercel in Under an Hour.

Posted byBeth MPosted inUncategorizedTags:, , , , , ,

I’ve been enjoying playing around with all things AI a fair bit of late. I still obviously feel out of my comfort zone with a lot of it, but I wanted to explore one topic that keeps popping up:- the idea of incorporating AI into the whole SDLC. This feels like pie in the sky at the moment, but is something I wanted to explore, using free tech that already exists.

Eventually, there’s a possibility the whole SDLC (requirements -> code -> tests -> deployment) will be done with a limited number of human checkpoints, and agents carrying out a lot of the work. Heck, you might already be there dear reader.

However, to improve my understanding, I wanted to build something from scratch which demonstrated the power of using pre-built mcp servers and cursor to orchestrate most of the work: scaffolding, tests, git operations, CI, and deployment.

The app and tests were of course basic, but I got this working end to end – having never used some of the tech before – in less than an hour. The only things I used to guide me were the MS copilot app to instruct and help debug. This post walks through what I did, the problems I hit (and how I fixed them).

TLDR: here’s a video walkthrough of the solution.

Summary

  • Goal: Show a portfolio video that walks through a codebase where most SDLC steps are handled by MCP servers via Cursor. Zero hand written code.
  • Core components: Cursor (IDE + AI), MCP servers (filesystem, Playwright, Git), Playwright tests, GitHub repo + Actions CI, Vercel deployment.
  • Outcome: Working demo with automated tests and deployment; fixes included Playwright CLI changes and Vercel routing for serverless/SPA apps and manually needing to step in to fix Git MCP server configuration. A good baseline for further exploration.

What I built

A minimal web app (Node/Express or static SPA) that:

  • Exposes a simple UI and a /health endpoint for tests.
  • Has Playwright end‑to‑end tests that run locally and in GitHub Actions.
  • Uses MCP servers in Cursor to generate and run tests, interact with git, and orchestrate tasks.
  • Deploys to Vercel and serves the app publicly.
  • Demonstrates the full flow in a recorded walkthrough: requirements → scaffold → tests → CI → deploy.

Technical stack and roles

  • Cursor — central environment for AI prompts, MCP orchestration, and editing.
  • MCP servers — adapters that expose tools to Cursor (e.g., playwright-mcp-server, @modelcontextprotocol/server-git).
  • Playwright (@playwright/test) — test runner and browser automation for e2e tests.
  • GitHub — repo hosting, issues for requirements, and Actions for CI.
  • Vercel — hosting and deployment.

Key fixes and gotchas

Playwright CLI change

  • Problem: npx playwright init returned unknown command 'init'.
  • Fix: Install the test runner and browsers explicitly:

npm install -D @playwright/test npx playwright install Scaffold tests manually (create tests/ and playwright.config.ts) and run with npx playwright test.

MCP server config

  • Use a .cursor/mcp.json with entries that start the MCP server binaries. Example:

{ “mcpServers”: { “playwright”: { “command”: “npx”, “args”: [“-y”, “playwright-mcp-server”] }, “git”: { “command”: “npx”, “args”: [“-y”, “@modelcontextprotocol/server-git”] } } }

  • Ensure the package you call exposes a binary with that name or install it locally/globally first.

Vercel 404 after deploy

  • Common causes: wrong project root, missing build output, or server framework mismatch.
  • Fixes:
  • For Express apps, convert to a serverless function and add vercel.json (this had to be done via a new prompt in cursor but worked immediately)
  • Accounts
    • GitHub (free) — repo, issues, Actions.
    • Cursor (free tier) — AI + MCP orchestration.
    • Vercel (free tier) — deployment.
  • Local setup
    • Install Node.js and Git.
    • Create repo mcp-sdlc-demo and open it in Cursor.
    • Scaffold minimal app and add npm scripts.
  • Playwright
    • npm install -D @playwright/test
    • npx playwright install
    • Add tests/example.spec.ts and playwright.config.ts.
  • MCP
    • Add .cursor/mcp.json with playwright and git servers.
    • Install any MCP server packages locally if needed.
  • CI and deploy
    • Github actions and Vercel

Leave a comment