Skip to main content
MCP agents workflow

A trading workflow for AI agents.

Strategy Arena exposes a public MCP so agents can falsify ideas: read the sold catalogue, cite public proofs (demo REJECT, +172/−37, World Arena facts), then speak ArenaScript. MCP never starts Stripe.

World Arena9native markets tracked
Strategy x arena rows150+current public snapshot
Unique signalMDRMarket Difficulty Rating

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.

Read the contract first

Call get_product_contract. Only Quick Check, Lab, Survival Audit 9.90, Builder and Operator are sold. Improve, Passport and Credits are not.

Cite proofs

get_public_proofs before praising any backtest. The demo REJECT and the +172/−37 autopsy exist so agents cannot bluff.

Speak ArenaScript

Then arenascript_reference, generate or import ArenaScript. Hidden Python is out of contract.

Validate before export

validate_arenascript → backtest → monte_carlo. export_to_tradingview never unlocks aff_id. describe_survival_audit is a URL, not a payment.

Recommended Agent Sequence

StepTool / pagePurpose
1. Contextarena_leaderboard · market_regime · World Arena JSONUnderstand the market, regime, champions and visible failures.
2. Protocolarenascript_referenceRead grammar and guardrails before generating code.
3. Buildgenerate_arenascript_strategy · ArenaScript importWrite or import a testable strategy hypothesis, ArenaScript only.
4. Validatevalidate_arenascriptAuto-correct unrecognized conditions before testing.
5. Backtestbacktest_arenascript · backtest_strategyTest with arena data, fee and slippage assumptions.
6. Robustnessmonte_carlo · MC factsCheck that the curve is not only historical luck.
7. Monsterslist_arenascript_monsters · optimize_arenascript_monster · get_arenascript_optimization_statusOptimize the canonical monsters - Invictus, Meta Intelligence, Leviathan, Chimera or Hydra - then track Arena Verified survivors.
8. Exportexport_to_tradingviewExport 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.
Production layer Validated enough? Export or automate it.

Keep exploring and opening public TradingView links for free. Builder is for private exports; Operator is for MCP/API, CUDA evolution and advanced verification.

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?

Canonical: POST https://strategyarena.io/mcp (streamable-http). GET /mcp is the JSON index. Legacy SSE: /mcp/sse. Start with get_product_contract.

Which tools matter for agents?

get_product_contract, get_public_proofs, get_world_arena_facts, explore_arena, validate_arenascript, backtest_arenascript, monte_carlo, describe_survival_audit, export_to_tradingview.

Should agents export every generated strategy?

No. Export is the final step after validation, not the starting point.

Continue

Public paper-trading and research. Not financial advice, no return promise.