Backtesting Arena — Agent Skill
Battle-tested market intelligence for AI agents — every claim backtested,
versioned, and explainable. Evidence and base rates, never financial advice.
Skill package for AI agents and LLMs that need programmatic access to the
Backtesting Arena API: daily Bitcoin /
crypto cycle scoring, on-chain indicators, aggregated strategy insights, and
on-demand backtests.
Three channels, one product:
| Channel | Auth | Use case |
|---|---|---|
| Public | none, fair-use cache | quick reads, no signup |
| REST + MCP | Bearer API key (Free tier available) | full surface, higher quotas |
| x402 pay-per-call | EIP-3009 wallet signature (USDC on Base) | autonomous agents, no account |
The full capability descriptor an agent should load is SKILL.md.
Install for Claude Code / Claude Desktop```
Coming soon when /install-skill GA's; for now clone + reference manually:
git clone https://github.com/Schoasch/skill-backtesting-arena.git
Reference `skills/backtesting-arena/SKILL.md` as a context document in your agent runtime, or use the
MCP server directly (see below).
## Quick start
### 1. Cycle-aware trading bot (Bearer)```bash
curl https://tradingstrategies.work/api/arena-pulse/today
# → { "score": 35, "band": "risk-off", ... } — no API key required
Want higher quotas + full surface? Get a free Bearer key.
2. Autonomous agent with USDC wallet (x402)```bash
First call returns HTTP 402 + payment instructions:
curl -i https://tradingstrategies.work/api/v1/agent/btc-cycle/latest
→ 402 Payment Required, body has accepts[] with payTo, asset, amount, network
Discovery descriptor lists all paid endpoints + pricing:
curl https://tradingstrategies.work/.well-known/x402
Full Python client showing the EIP-3009 sign + retry pattern: `examples/python-x402-client.py`.
### 3. MCP server (Claude Desktop, Cursor, etc.)
Two transports — HTTP (recommended, requires Bearer key) or stdio.
HTTP variant (drop into `claude_desktop_config.json` → `mcpServers`):```json
{
"backtesting-arena": {
"url": "https://tradingstrategies.work/api/mcp",
"headers": { "Authorization": "Bearer sk-arena-…" }
}
}
Full example incl. stdio variant + Cursor config: examples/claude-desktop-config.json.
Examples folder
| File | What |
|---|---|
examples/curl-snippets.sh |
Bash snippets for public + Bearer endpoints |
examples/python-x402-client.py |
Python EIP-3009 signing + retry for x402 endpoints |
examples/typescript-mcp-client.ts |
Connect to MCP server, list the 49 default tools, call one |
examples/claude-desktop-config.json |
Drop-in config for Claude Desktop / Cursor |
Discovery files (machine-readable)
For AI tooling that auto-discovers APIs:
/openapi.json— full OpenAPI 3.1 spec, ~70 paths,x-x402extension on agent endpoints/skill.md— this skill (canonical, kept in sync with this repo)/llms.txt— LLM-friendly summary/.well-known/api.json— cross-tool discovery hub/.well-known/x402— Bazaar-v2 descriptor of paid endpoints (alias.json)/.well-known/ai-plugin.json— OpenAI-style plugin manifest
Pricing
- Free — public endpoints + Free Bearer tier (30 req/h, 200/d)
- API Pro €9,99/mo — 300 req/h, 3k/d, backtest triggers, larger windows
- API Power €29,99/mo — 2000 req/h, 30k/d, universe-backtests (50 pairs)
- x402 pay-per-call — $0.01-$0.50 USDC per call, no subscription
One subscription covers both REST and MCP channels. Pricing is reviewed
quarterly against agent-economy benchmarks (changelog).
License
MIT (this skill package). The Backtesting Arena API itself follows the
Terms.
Operator
Solo-built crypto strategy backtesting platform from Germany/Austria. Public
since 2026. Educational and research focus, not signal-selling.
Contact: info@tradingstrategies.work · Issues: github.com/Schoasch/skill-backtesting-arena/issues
