MCP agents workflow
A trading workflow for AI agents.
Strategy Arena is not only a site humans read. It exposes a public MCP server so agents can speak ArenaScript, validate their own output, send monsters to CUDA and hand off verified systems to TradingView.
Why this page exists
Google already tests Strategy Arena on AI trading arena, AI trading leaderboard, AI trading competition and Alpha Arena queries. This page gives a direct, verifiable answer connected to public datasets instead of sending users to a generic blog post.
Speak ArenaScript
Agents should start with arenascript_reference, then generate or import ArenaScript instead of writing hidden Python or vague trading prose.
Validate before export
A generated strategy is not an edge. It should pass validate_arenascript, backtest_arenascript, monte_carlo and methodology checks before export_to_tradingview is used.
Optimize monsters
When working on the canonical monsters - Invictus, Meta Intelligence, Leviathan, Chimera or Hydra - agents can use list_arenascript_monsters, optimize_arenascript_monster and get_arenascript_optimization_status. Edge Fund, ML gates, GPU War and ScalpForge are ArenaScript systems, not monsters.
Keep caveats visible
All MCP outputs are paper-trading research. Agents should preserve caveats around fees, slippage, drawdown, Buy & Hold and no financial advice.
Recommended Agent Sequence
| Step | Tool / page | Purpose |
|---|---|---|
| 1. Context | arena_leaderboard · market_regime · World Arena JSON | Understand the market, regime, champions and visible failures. |
| 2. Protocol | arenascript_reference | Read grammar and guardrails before generating code. |
| 3. Build | generate_arenascript_strategy · ArenaScript import | Write or import a testable strategy hypothesis, ArenaScript only. |
| 4. Validate | validate_arenascript | Auto-correct unrecognized conditions before testing. |
| 5. Backtest | backtest_arenascript · backtest_strategy | Test with arena data, fee and slippage assumptions. |
| 6. Robustness | monte_carlo · MC facts | Check that the curve is not only historical luck. |
| 7. Monsters | list_arenascript_monsters · optimize_arenascript_monster · get_arenascript_optimization_status | Optimize the canonical monsters - Invictus, Meta Intelligence, Leviathan, Chimera or Hydra - then track Arena Verified survivors. |
| 8. Export | export_to_tradingview | Export Pine v5 only after validation, never as the first step. |
MCP Prompt
Any AI can speak ArenaScript
generate · validate · backtest · CUDA · export
Prompt for Grok / Claude
Connect to https://strategyarena.io/mcp/sse. Call arenascript_reference. Generate ArenaScript for [idea]. Validate with validate_arenascript until clean. Backtest with backtest_arenascript. Run monte_carlo. If it is a monster, optimize_arenascript_monster and poll get_arenascript_optimization_status. Export only after validation.
1. SpeakThe agent speaks ArenaScript, not hidden Python or vague prose.
2. VerifyValidate, backtest, Monte Carlo, Hospital and Buy & Hold filter fragile ideas.
3. EvolveCUDA optimizes monsters, then the public log keeps Arena Verified survivors.
4. ExportTradingView is the production handoff, never the shortcut replacing validation.
FAQ
What MCP endpoint does Strategy Arena expose?
The public SSE endpoint is https://strategyarena.io/mcp/sse.
Which tools matter for agents?
arenascript_reference, generate_arenascript_strategy, validate_arenascript, backtest_arenascript, monte_carlo, optimize_arenascript_monster, get_arenascript_optimization_status, get_monster_evolution_log and export_to_tradingview are the core workflow tools.
Should agents export every generated strategy?
No. Export is the final step after validation, not the starting point.
Continue
World Arena
World Arena facts
Strategy Hospital
Methodology
TradingView validation
Validated strategies
Pine converter
Calibration
GPU Arena
Public paper-trading and research. Not financial advice, no return promise.