MCP server for Golden Code methodology - everything you vibecode turns to gold
npm install midas-mcp

Everything you vibecode turns to gold.
An MCP server that brings the Golden Code methodology into Cursor as an interactive coach. Guides users through the two-phase process, provides expert prompts, and audits projects against the 12 ingredients of production readiness.
Midas uses a hybrid architecture for optimal performance:
| Component | Purpose | Setup |
|-----------|---------|-------|
| Claude Code | Executes prompts, makes code changes | curl -fsSL https://claude.ai/install.sh \| bash then claude to auth |
| Anthropic API Key | Fast project analysis, smart suggestions | Add to ~/.midas/config.json (optional but recommended) |
``bash1. Install Claude Code (required for auto-mode)
curl -fsSL https://claude.ai/install.sh | bash
Midas will guide you through any remaining setup (GitHub login, API key).
$3
- Claude Code = "The Hands" - actually modifies files, runs commands, iterates on errors
- Anthropic API = "The Brain" - fast analysis (~$0.003/analysis), generates smart suggestions
Without Claude Code, Midas can only analyze. With Claude Code, Midas can execute prompts and grind through your gameplan in auto-mode.
Quick Start
Run Midas instantly (no install required):
`bash
npx --yes midas-mcp@latest
`This launches the interactive TUI coach. Press
? for help, p to copy the recommended prompt.$3
`bash
npx midas-mcp run
`This does everything:
1. Checks Claude Code installation and auth
2. Logs you into GitHub (for cloud dashboard)
3. Syncs your project to the dashboard
4. Starts watch mode for remote/auto execution
Installation (Optional)
For global installation (enables
midas command):`bash
npm install -g midas-mcp
`Or add a shell alias to
~/.zshrc or ~/.bashrc:`bash
alias midas="npx midas-mcp@latest"
`Then reload your shell:
source ~/.zshrcCursor Configuration
Add to your Cursor settings (
~/.cursor/mcp.json):`json
{
"mcpServers": {
"midas": {
"command": "npx",
"args": ["midas-mcp@latest", "server"]
}
}
}
`Using
@latest ensures you always get the newest version. The server argument runs midas as an MCP server (without it, you get the interactive TUI).What Midas Does
$3
Midas tracks your current phase and guides you through:
Phase 1: Plan (Pre-Build)
`
π‘ Idea β π Research β π PRD β πΊοΈ Gameplan
`Phase 2: Build (7-Step Process)
`
1. User Rules β 2. Index Codebase β 3. Read Files β 4. Research
5. Write Code + Tests β 6. Run Tests β 7. Log β Fix β Repeat
`$3
| Tool | Description |
|------|-------------|
|
midas_start_project | Initialize Plan, create docs folder structure |
| midas_get_phase | Get current phase and recommended next steps |
| midas_set_phase | Move to a specific phase |
| midas_audit | Score project against 12 ingredients |
| midas_check_docs | Verify prd/gameplan completeness |
| midas_oneshot | Apply the Oneshot Paradigm for error recovery |
| midas_tornado | Trigger Research + Logs + Tests cycle |
| midas_horizon | Expand horizontal context for better output |$3
Pre-built expert prompts you can invoke:
| Prompt | Purpose |
|--------|---------|
|
master_prompt | Session initialization with full context loading |
| execution_prompt | Start building with TDD approach |
| safe_refactor | Move code safely between files |
| bug_investigation | Root cause analysis before fixing |
| security_review | Check for vulnerabilities |
| feature_planning | Plan before implementing |$3
Documentation available on demand:
-
midas://methodology - Core philosophy
- midas://plan - Pre-build process
- midas://process - 7-step development loop
- midas://oneshot - Fix backward paradigm
- midas://tornado - Research + Logs + Tests
- midas://horizon - Horizontal vs vertical context
- midas://ingredients/{1-12} - Production guidelines
- midas://cheatsheet - Quick referenceThe Three Principles
When you hit problems, Midas applies:
$3
When things break, go back with error + context instead of patching forward.
`
Original Prompt + Error Log + "Avoid this" = Working Solution
`$3
Three forces spinning together solve any problem:
`
Research + Logs + Tests = Solution
`$3
AI thinks vertical (implementation). You provide horizontal (context).
`
Wrong output? Widen your horizontal context.
`Usage Examples
$3
`
User: "I want to build a task management app"Midas: Initiates Plan phase, creates docs folder,
guides through PRD β Gameplan
`$3
`
User: "This keeps failing with the same error"Midas: Detects repeated failures, suggests Oneshot approach,
helps construct enhanced prompt with error context
`$3
`
User: "Is my project ready to ship?"Midas: Runs 12 ingredients audit, scores each category,
identifies gaps and provides remediation guidance
`The 12 Ingredients
Midas audits against:
| # | Ingredient | Category |
|---|------------|----------|
| 1 | Frontend | Core |
| 2 | Backend | Core |
| 3 | Database | Core |
| 4 | Authentication | Core |
| 5 | API Integrations | Power |
| 6 | State Management | Power |
| 7 | Design/UX | Power |
| 8 | Testing | Protection |
| 9 | Security | Protection |
| 10 | Error Handling | Protection |
| 11 | Version Control | Mastery |
| 12 | Deployment | Mastery |
Completion levels: 1-4 (functional), 1-8 (polished), 1-12 (production-ready)
Cloud Dashboard
Track your progress across all projects at dashboard.midasmcp.com.
$3
`bash
npx midas-mcp login
`This opens GitHub for authentication. Once complete, your projects can sync to the cloud.
$3
`bash
cd your-project
npx midas-mcp sync
`Syncs current phase, gate status, and recent activity to the dashboard.
$3
`bash
npx midas-mcp whoami
`Shows your authenticated GitHub username and sync configuration status.
Remote Pilot (Control from Phone)
Run AI coding tasks from your phone while your laptop works in the background.
$3
`bash
npx midas-mcp pilot --remote
`This displays a QR code in your terminal. Scan it with your phone to open the mobile dashboard.
$3
1. Laptop: Run
npx midas-mcp pilot --remote - displays QR code, waits for commands
2. Phone: Scan QR β opens mobile dashboard at dashboard.midasmcp.com/pilot/...
3. Execute: Tap quick actions or type custom prompts
4. Claude Code runs locally on your laptop, output streams to your phone$3
- Run Tests: Execute test suite and fix failures
- Fix Lints: Run linter and auto-fix issues
- Build: Compile project and resolve errors
- Analyze: Get AI analysis and next step suggestions
- Custom prompts: Send any instruction to Claude Code
$3
- Claude Code CLI installed on your laptop
- GitHub account (for authentication)
$3
`bash
Watch mode - poll dashboard for queued commands
npx midas-mcp pilot --watchSingle command execution
npx midas-mcp pilot "Fix the bug in auth.ts"
`Development
`bash
cd midas-mcp
npm install
npm run build
npm run dev
``MIT