⚡ MegaPad CLI & Native MCP Server

Run Claude & ChatGPT Side-by-Side with Zero API Keys

Scientific Multi-Model Benchmarking, Instant Diff Inspection & Native MCP Server for Claude Code, Cursor, and Terminal.

Quickstart · See It in Action · Interactive Actions · Native MCP · Zero-Key Auth · Contributing

⚡ What is MegaPad?

Stop copy-pasting code between browser tabs. MegaPad automatically bridges your existing Claude Code, ChatGPT Plus / Codex, and Grok CLI logins into one unified local control plane.

• Run Frontier Models in Parallel: Race Claude, GPT-6 Astra, DeepSeek V4.1, Grok, and Gemini on the same prompt.

• Zero API Keys Needed: Auto-detects and bridges your active flat-rate subscriptions on disk. No metered API charges.

• Interactive 1-Keystroke Actions: Copy the winner's code to your clipboard ([1] or [2]), inspect a colored terminal diff ([D]), or write directly to a file ([W]).

• Native MCP Server: Turn Claude Code and Cursor into multi-model agents that automatically consult rival models over stdio.

Related MCP server: RespCode MCP Server

🖥️ See It in Action

Run a 1-on-1 head-to-head race:

pad vs openai "Write a lock-free ring buffer in Rust"

┌────────────────────────────────────────────────────────────────────────────────────────┐
│  megapad — head-to-head duel · PARALLEL EXECUTION                                      │
├────────────────────────────────────────────────────────────────────────────────────────┤
│ $ pad vs openai "Write a lock-free ring buffer in Rust"                                │
│                                                                                        │
│ ⚡ Claude Fable 5.1 (1,140ms · 110 tok/s)    ⚡ OpenAI GPT-6 Astra (1,020ms · 130 tok/s)│
│ ───────────────────────────────────────     ────────────────────────────────────────── │
│ pub struct RingBuffer<T> {                  pub struct LockFreeQueue<T> {              │
│     buffer: Box<[UnsafeCell<T>]>,               buffer: Vec<AtomicPtr<T>>,             │
│     head: AtomicUsize,                          read_idx: CachePadded<AtomicUsize>,    │
│     tail: AtomicUsize,                          write_idx: CachePadded<AtomicUsize>,   │
│ }                                           }                                          │
├────────────────────────────────────────────────────────────────────────────────────────┤
│ Telemetry: GPT-6 Astra was 120ms faster (130 vs 110 tok/s). Total Cost: $0.00000       │
│                                                                                        │
│ ⚡ ACTIONS (100% offline · 0 additional tokens):                                        │
│   [1] Copy Claude code to clipboard   [D] View colored unified diff                     │
│   [2] Copy OpenAI code to clipboard   [W] Write winning code to file                    │
│   [Q / Enter] Done (exit)                                                              │
└────────────────────────────────────────────────────────────────────────────────────────┘

⚡ Quickstart

1. Install Globally or Run with NPX

npm install -g megapad

# Or zero-install via npx:
npx megapad "How do I optimize this Postgres query?"

2. Verify Your Connected Sessions

pad status

🔑 MEGAPAD CONNECTED SESSIONS & ZERO-KEY FRONTIER MODELS
✔ Claude Code: (~/.claude.json) · Active: Claude Fable 5.1
✔ Codex / ChatGPT: (~/.codex/auth.json) · Active: GPT-6 Astra
✔ Grok CLI: (~/.grok/auth.json) · Active: Grok 4.6
⚡ Google Gemini: (Google Cloud Free Tier) · Active: Gemini 3.8 Flash

3. Run Duels, Races & Peer Reviews

# Head-to-head duel (Claude vs Target):
pad vs openai "Implement an atomic rate limiter"
pad vs grok "Find edge-case race conditions in this concurrency loop"
pad vs deepseek "Refactor this callback hell to async/await"

# Multi-model parallel race across all engines:
pad "How do I fix memory leaks in Node.js event emitters?"

# Peer code review council of staged git changes:
git diff --staged | pad review

🎮 Interactive Post-Duel Actions (Zero Extra Tokens)

After every race, MegaPad lets you take action with 1 keystroke:

• [1] or [2] (Copy Code to Clipboard): Automatically strips conversational markdown fluff, extracts the core implementation, and places it straight on your clipboard (pbcopy / clip / wl-copy). Ready to Cmd+V into your editor.

• [D] (Colored Unified Diff): Computes a git-style line diff (+ green / - red) directly in your terminal so you can immediately see the architectural trade-offs.

• [W] (Write to File): Enter a destination path (e.g. src/buffer.rs) to write the winning solution directly to disk.

• [Q / Enter]: Exits cleanly. Automatically skips prompting when running in pipes or scripts.

🤖 Claude Code & Cursor Integration (Native MCP)

Turn your AI coding agent into a multi-model orchestrator.

1-Click Auto Configurator

pad install
# or via npx:
npx mechapad install

Auto-detects and configures Claude Code (~/.claude.json), Cursor (~/.cursor/mcp.json), and Codex CLI.

MCP Tools Provided:

• megapad_compare: Ask multiple frontier models the same prompt in parallel and return structured benchmark metrics.

• megapad_consult_peer: Ask a rival model (e.g. GPT-6 or Grok) for an immediate second opinion when Claude gets stuck.

• megapad_council_code_review: Sends code or diffs to a multi-model council (OpenAI + DeepSeek) to catch edge-case bugs and security vulnerabilities.

🔑 Zero-Key Local Session Bridge

MegaPad never asks you to paste paid API keys:

1. Local Disk Credentials: Reads existing authorized CLI session credentials from ~/.claude.json, ~/.codex/auth.json, and ~/.grok/auth.json.

2. 100% Local Stdio: Runs on localhost over standard stdio. No telemetry, no third-party proxies, zero code leaves your machine.

3. No Token Double-Billing: Queries execute through your existing flat-rate subscriptions (ChatGPT Plus, Claude Pro, Grok). Routine tasks route to Gemini Flash (100% free forever) or DeepSeek for pennies.

📊 Scientific Benchmark Telemetry

Every query returns live, high-precision latency, tokens per second, and estimated cost:

🤝 Contributing

We welcome community contributions! Want to add a new model provider (Ollama, Qwen, Mistral, Bedrock) or support a new IDE? Check out our Contributing Guide to get started in 10 minutes.

Good First Issues

Check out issues with the good first issue label to contribute!

📈 Star History

📄 License

MIT © Eugene Finch & MegaPad Contributors