Skip to content

MCP-SETUP.md

Latest commit

 

History

History
70 lines (45 loc) · 2.76 KB

MCP-SETUP.md

File metadata and controls

70 lines (45 loc) · 2.76 KB

MCP Server Configuration Guide

This project ships with 4 pre-configured MCP (Model Context Protocol) servers that dramatically extend what the AI can do inside your project.

What is MCP?

MCP lets your AI coding assistant (Cursor, Claude) interact directly with external tools — reading GitHub PRs, querying your database schema, browsing web pages, and accessing your local filesystem — all from within the chat window.

This is the #1 feature that separates Vibe Stack from every other boilerplate. Other templates give you code. We give you a fully wired AI development environment.

Server Configuration

1. GitHub MCP Server

Purpose: Lets the AI read your GitHub repos, PRs, issues, and commits directly.

Setup:

  1. Go to github.com/settings/tokens
  2. Create a Fine-grained personal access token with read access to your repos
  3. Replace <YOUR_GITHUB_PAT> in .cursor/mcp.json

Example Use:

"Read the open pull request on the vibe-stack repo and review the code changes for security issues based on our .mdc rules."

2. Filesystem MCP Server

Purpose: Gives the AI read/write access to your project directory.

Setup: Works out of the box — no configuration needed.

Example Use:

"List all files in the src/app/ directory and identify any components that are missing error boundaries."

3. Supabase MCP Server ⭐ (Vault Feature)

Purpose: Lets the AI query your Supabase project directly — inspect tables, read RLS policies, and validate schemas.

Setup:

  1. Go to supabase.com/dashboard/account/tokens
  2. Generate an access token
  3. Replace <YOUR_SUPABASE_PAT> in .cursor/mcp.json

⚠️ Security: The server is configured with --read-only by default. The AI can inspect your schema but cannot modify your production data.

Example Use:

"Connect to my Supabase project and check if the profiles table has RLS enabled. If not, generate the policy SQL."

4. Browser MCP Server ⭐ (Vault Feature)

Purpose: Lets the AI control a headless browser — navigate pages, take screenshots, fill forms, and extract data.

Setup: Works out of the box if you have Node.js installed (uses Puppeteer).

Example Use:

"Open localhost:3000/dashboard in the browser, take a screenshot, and tell me if the layout matches our design spec."

Troubleshooting

Issue Fix
Green dot doesn't appear in Cursor Restart Cursor after saving mcp.json
"ENOENT" error Run npm install to ensure npx is available
GitHub rate limiting Upgrade to a GitHub Pro token or reduce request frequency
Supabase connection refused Verify your project URL and access token are correct