Your AI, your GPU
1,000,000 strategy variants in about 20 seconds on an RTX 4080 Super (Lab 235 measurement, 7 October 2026, legacy execution semantics). Every row that is kept is replayed on the CPU reference engine.
Strategy Arena Lab · MCP
Claude, GPT, Gemini, Grok, DeepSeek or a model running on your PC can drive the Strategy Arena Lab: write a rule, backtest it, search thousands of variants on your graphics card, export the Pine for TradingView. Then the judge replays it on data nobody has seen and answers: winner or mirage?
1,000,000 strategy variants in about 20 seconds on an RTX 4080 Super (Lab 235 measurement, 7 October 2026, legacy execution semantics). Every row that is kept is replayed on the CPU reference engine.
Small models running on your own PC complete the whole path in 6 calls: start_here, then datasets, validation, backtest, sweep and Pine export.
An AI is very good at finding a rule that worked in the past. The sealed proof keeps a test period aside, compares with Buy & Hold on the same windows and counts every variant tried, so luck is not mistaken for skill.
The Lab speaks MCP, the protocol that lets an AI use software. Your AI client starts a small local script; the script talks to the Lab running on your PC.
Requirement: Node.js 20.3 or newer for the connection script. Replace PATH_COPIED_FROM_THE_LAB with the path the Lab gives you.
File: %APPDATA%\Claude\claude_desktop_config.json (Windows).
{
"mcpServers": {
"strategyarena-lab": {
"command": "node",
"args": ["PATH_COPIED_FROM_THE_LAB"]
}
}
}
File: ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project).
{
"mcpServers": {
"strategyarena-lab": {
"command": "node",
"args": ["PATH_COPIED_FROM_THE_LAB"]
}
}
}
In LM Studio: Program → Install → Edit mcp.json. Pick a model that supports tool calls.
{
"mcpServers": {
"strategyarena-lab": {
"command": "node",
"args": ["PATH_COPIED_FROM_THE_LAB"]
}
}
}
File: opencode.json in your project (or ~/.config/opencode/opencode.json).
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"strategyarena-lab": {
"type": "local",
"command": ["node", "PATH_COPIED_FROM_THE_LAB"],
"enabled": true
}
}
}
claude mcp add strategyarena-lab -- node "PATH_COPIED_FROM_THE_LAB"
By default the Lab shows a short list of tools with compact answers, so that small models can drive it too. start_here gives the whole path with ready-to-send arguments.
Need every tool? Add "--profile", "full" after the script path in args (or set STRATEGY_ARENA_MCP_PROFILE=full), then restart the client.
TradingView's Strategy Tester is the author's tool: it shows what the author chose to test, with no sealed period. The Lab acts as the second opinion. It exports the exact Pine, you paste TradingView's List of Trades back, and compare_tradingview_trades checks every trade.
6 of the 6 leaderboard strategies, 234 of 234 trades, give the same trades one by one in the real TradingView, quantities included (Lab 236 export, test of 8 October 2026).
Open TradingView (partner link). Strategy Arena may earn a commission on an eligible purchase; a free account is enough to compare trades. It never changes a verdict.
| Free | Operator 29 EUR/month | |
|---|---|---|
| MCP connection, essential profile, backtest, sweeps, Pine export | ✓ | ✓ |
| Sealed proof “Winner or mirage?” | ✓ | ✓ |
| TradingView comparisons | limited | unlimited |
| Background GPU jobs above 200,000 variants, Think above 100,000 | — | ✓ |
| Night mode and multi-asset jobs | — | ✓ |
| Full proof history | — | ✓ (Builder and up) |
Paying never changes a verdict.
When a request needs Operator, the Lab tells your AI once, in one sentence. The AI never buys anything.
Launch access is not open yet. When it opens, every account created or signed in during the launch window gets full Operator access in the Lab for free, without a card, until a fixed end date announced here.
Founder member offer: accounts created during the launch (opening planned for 15 October 2026) can then keep Operator at 19 €/month instead of 29 €/month, for as long as the subscription stays active.
The Lab is local and computes on your machine. The site also offers an online MCP server (https://strategyarena.io/mcp, Streamable HTTP) to read the public catalogue and proofs: MCP for agents.
Educational research and backtesting tool. Past results, not a forecast. No investment advice.