MegaPad — Run Claude & ChatGPT Side-by-Side

Q: How does MegaPad run Claude, ChatGPT, and Grok without API keys?

MegaPad automatically detects and bridges your existing local CLI session credentials from ~/.claude.json, ~/.codex/auth.json, and ~/.grok/auth.json. It uses your active flat-rate subscriptions rather than charging per-token metered API rates.

Q: Which AI models can I run side-by-side?

MegaPad supports Claude (Fable 5.1, Sonnet 5), OpenAI (GPT-6 Astra, Sol, 4o), Grok (4.6, Build), Kimi (K2.7 Code / TC), DeepSeek (V4.1 Flash, R1), and Gemini (3.8 Flash, Free Tier). You can race models in parallel or duel them 1-on-1.

See It in Action

One prompt. Two frontier engines. Instant side-by-side verification.

Both models run concurrently against your exact prompt. Compare architectures, inspect microsecond latency, token throughput, and exact cost side-by-side in your terminal.

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

🔬 MEGAPAD SCIENTIFIC MULTI-MODEL BENCHMARK
Prompt: "Write a lock-free ring buffer in Rust"
Target Providers: claude, openai
⏳ Parallel execution complete. Live telemetry:

┌──────────────────────────┬────────┬──────────┬──────────┬──────────┬───────────┬─────────────┬──────────────────┐
│ Provider / Model         │ Status │ Latency  │ In / Out │ Total    │ Speed     │ Cost (USD)  │ vs Claude Cost   │
├──────────────────────────┼────────┼──────────┼──────────┼──────────┼───────────┼─────────────┼──────────────────┤
│ OpenAI GPT-6 Astra       │ ✔ OK   │ 1,020ms  │ 120/510  │ 630      │ 130 tok/s │ $0.00000    │ $0 Extra (Codex) │
│ Claude Fable 5.1         │ ✔ OK   │ 1,140ms  │ 120/480  │ 600      │ 110 tok/s │ $0.00000    │ Base (100%)      │
└──────────────────────────┴────────┴──────────┴──────────┴──────────┴───────────┴─────────────┴──────────────────┘
Total Parallel Wall Clock Time: 1,142ms

[OpenAI GPT-6 Astra] (1,020ms · 630 tokens · 130 tok/s · $0.00000)

────────────────────────────────────────────────────────────────────────────────

pub struct LockFreeQueue<T> {
    buffer: Vec<AtomicPtr<T>>,
    read_idx: CachePadded<AtomicUsize>,
    write_idx: CachePadded<AtomicUsize>,
}
// Cache-padded atomic indices prevent false sharing between CPU cores

[Claude Fable 5.1] (1,140ms · 600 tokens · 110 tok/s · $0.00000)

────────────────────────────────────────────────────────────────────────────────

pub struct RingBuffer<T> {
    buffer: Box<[UnsafeCell<MaybeUninit<T>>]>,
    head: AtomicUsize,
    tail: AtomicUsize,
    capacity_mask: usize,
}
// CAS-free atomic ring buffer using power-of-two masking

Get Started in 30 Seconds

Three simple steps. Zero configuration headaches.

Just like Caveman and CodeGraph, MegaPad is built to get out of your way and start working in seconds.

1

Install the CLI & MCP Server

Install MegaPad globally via npm, or run directly using npx. Everything is self-contained with zero runtime dependencies.

npm install -g megapad # or wire up your IDEs in 1 step: npx mechapad install

2

Verify Your Local Sessions

Run pad status. MegaPad automatically discovers existing login sessions from your disk. No API key copy-pasting required.

$ pad status ✓ Claude Code: (~/.claude.json) ✓ Codex / ChatGPT: (~/.codex/auth.json) ✓ Grok CLI: (~/.grok/auth.json) ✓ Gemini 3.8 Flash (Free Forever)

3

Duel or Peer-Review Anywhere

Run head-to-head duels from your terminal, or ask Claude inside Claude Code / Cursor to consult GPT-6 via native MCP tools.

# Run a 1-on-1 duel: pad vs openai "Fix concurrency deadlock" # Or peer-review current git diff: git diff | pad review

Flexible Developer Workflow

Two ways to use MegaPad. Terminal or inside your editor.

Whether you love fast terminal commands or want Claude Code and Cursor to consult rival models automatically, MegaPad fits your workflow.

💻 1. The Terminal CLI (`pad`)

Lightweight, blazing-fast terminal companion for side-by-side model races, granular duels, and git diff peer reviews.

Parallel Races: pad "How do I optimize this Postgres query?" executes across all available engines at once.
Head-to-Head Duels: pad vs grok "prompt" or pad vs deepseek "prompt" pits your primary model against a specific rival.
Peer Review Council: git diff | pad review sends your staged git changes to a council of OpenAI + DeepSeek for edge cases.
Scientific Telemetry: Microsecond latency, tokens/sec throughput, and token-cost breakdown printed automatically for every run.

🔌 2. Native MCP in Claude Code & Cursor

Turn your active AI coding agent into a multi-model orchestrator. Auto-configured over stdio with zero background daemons.

1-Command Setup: npx mechapad install detects and configures Claude Code and Cursor automatically.
Instant Second Opinions: When Claude gets stuck, ask Claude: "Ask GPT-6 for alternative approaches" — Claude invokes megapad_consult_peer.
Council Code Reviews: Claude invokes megapad_council_code_review to find security vulnerabilities and memory bugs before applying edits.
Zero Daemon Leaks: Spawns cleanly over stdio when your agent runs, and terminates when it exits. No battery drain.

Zero API Key Architecture

How local session auto-bridging actually works.

You already pay $20/month for AI subscriptions. MegaPad makes sure you never pay twice for API tokens.

1. Local Disk Discovery

When you log into the official Claude Code CLI, Codex CLI, or Grok CLI, your authenticated session tokens are stored locally on disk (in ~/.claude.json, ~/.codex/auth.json, and ~/.grok/auth.json).

2. 100% Local Stdio Bridge

MegaPad reads those credentials directly on localhost. Prompts and responses stream straight between your terminal and the official endpoints over stdio. No third-party servers, no middleman proxy.

3. Zero Metered Token Bills

Because your queries run through your existing flat-rate subscriptions (ChatGPT Plus, Claude Pro, Grok), you don't burn through costly API credit cards. Routine tasks route to Gemini Flash (100% free) or DeepSeek for pennies.

Scientific Telemetry

Stop guessing. Measure tokens, latency & cost live.

Every time you run a query, MegaPad measures microsecond latency, input/output token counts, exact generation speed in tokens/second, and calculates exact dollar cost vs Claude Fable 5.1.

Provider / Model	Status	Latency	Speed	Per-Query Cost	vs Claude Cost
Claude Fable 5.1 / Sonnet 5	Active Session	1,140ms	110 tok/s	$0.00000	$0 Extra (Claude Sub)
OpenAI GPT-6 Astra / Sol	Active Session	1,020ms	130 tok/s	$0.00000	$0 Extra (Codex Sub)
Grok 4.6 / Build	Active Session	1,620ms	69 tok/s	$0.00000	$0 Extra (Grok Sub)
Kimi K2.7 Code / TC	Active Cloud	480ms	195 tok/s	$0.00011	98% Cheaper
DeepSeek V4.1 Flash / R1	Active Cloud	689ms	185 tok/s	$0.00010	97% Cheaper
Gemini 3.8 Flash	Free Forever	940ms	145 tok/s	$0.00000	100% Free ($0)

Native Model Context Protocol

Add Multi-Model Superpowers to Claude Code & Cursor

Zero background daemons to manage. Wire MegaPad into your active AI coding agent in one command.

1. 1-Command Auto-Configurator

One command scans and auto-configures Claude Code, Cursor, and Codex CLI:

npx mechapad install

2. Instant MCP Tool Suite

Provides megapad_compare, megapad_consult_peer, and megapad_council_code_review tools inside your active editor.

3. Clean Stdio Lifecycle

Runs cleanly over stdio. Starts when your AI agent runs, stops when it exits. No persistent memory leaks, no background batteries drained.

The Cost Slasher

Stop paying $150/mo for multiple AI subscriptions.

MegaPad routes your routine coding tasks to free and ultra-cheap models (DeepSeek, Gemini Free Tier), reserving expensive frontier models only for complex architectural decisions.

Standalone Subscriptions

$100/mo

ChatGPT Plus ($20) + Claude Pro ($20) + Cursor Pro ($20) + Grok ($40)

Separate windows & copy-paste fatigue
No side-by-side verification
Expensive token billing

MegaPad CLI & MCP

$0/free

Open-source software. Run models concurrently on your machine.

Claude Code & Cursor native MCP
Zero API key lock-in (Local sessions)
Up to 98% cheaper via DeepSeek & Gemini
Microsecond latency & token telemetry

Install via NPX

Frequently Asked Questions

Everything you need to know.

Straight answers on local session auth, MCP integration, and multi-model workflows.

How does MegaPad run models without API keys?

MegaPad automatically reads session credentials from your local configuration files (like ~/.claude.json, ~/.codex/auth.json, and ~/.grok/auth.json). You use the flat-rate subscriptions you already pay for instead of paying twice for API tokens.

How is this different from copying between browser tabs?

MegaPad executes prompts across frontier models concurrently with zero manual copying. You get exact microsecond latency, tokens/second speed benchmarks, and full code responses side-by-side in your terminal or IDE without context switching.

Can I use MegaPad inside Claude Code and Cursor?

Yes. MegaPad is a Native Model Context Protocol (MCP) server. Running npx mechapad install configures your Claude Code and Cursor environments to provide multi-model compare, peer consultation, and review tools directly inside your editor.

Which AI models can I run side-by-side?

MegaPad natively supports Claude (Fable 5.1, Sonnet 5), OpenAI (GPT-6 Astra, Sol, 4o), Grok (4.6, Build), Kimi (K2.7 Code / TC), DeepSeek (V4.1 Flash, R1), and Gemini (3.8 Flash, Free Tier). You can race models in parallel or duel them 1-on-1 with live telemetry.

Does MegaPad require any API keys or paid credits?

No. MegaPad auto-bridges your existing local session credentials from ~/.claude.json, ~/.codex/auth.json, and ~/.grok/auth.json. You use the subscriptions you already pay for. Routine edits can also route to Gemini 3.8 Flash (100% free forever) or local DeepSeek via Ollama for zero-cost operation.

Is my code sent to any third-party MegaPad servers?

No. MegaPad is 100% open-source and runs entirely on your local machine. Prompts and code are sent only to official model endpoints using your own authenticated sessions over stdio.

Get Started in Seconds

Start racing models in 5 seconds.

No signup. No credit card. Install globally and run your first multi-model duel.

$ npm install -g megapad && pad status

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