unCoded
Model Context Protocol

13,000+ Crypto Backtests, Queryable by Your AI

unCoded runs a public, read-only MCP server over its entire backtesting dataset. Connect Claude, ChatGPT or Cursor and ask questions about real strategy performance. Free, no API key.

The Endpoint

https://uncoded.ch/api/mcp

Transport: Streamable HTTP · Read-only · No authentication

Connect Your Client

Claude Code

claude mcp add --transport http uncoded https://uncoded.ch/api/mcp

Claude.ai & Claude Desktop

Settings → Connectors → Add custom connector, then paste the endpoint URL.

Cursor

{
  "mcpServers": {
    "uncoded": { "url": "https://uncoded.ch/api/mcp" }
  }
}

Any stdio client

Clients without native HTTP support can bridge via mcp-remote:

npx mcp-remote https://uncoded.ch/api/mcp

Seven Tools, One Dataset

list_symbols_and_modes

List symbols and strategy modes

All trading pairs and strategy modes in the dataset, with backtest counts. The usual starting point for an agent.

dataset_stats

Dataset statistics

Dataset-wide numbers: total backtests, snapshot age, share of strategies that beat buy-and-hold.

search_backtests

Search backtests

Filter by symbol, mode, return, win rate, drawdown or trade count. Sorted and paginated.

get_top_performers

Top performing backtests

The best results by return, alpha or Sharpe ratio, overall or per trading pair.

compare_strategy_vs_buyhold

Compare strategy vs buy & hold

How a strategy performed against simply holding the token over the same time window.

symbol_summary

Symbol summary

Aggregated view of one trading pair: result distribution, best modes, typical drawdowns.

get_backtest

Get a single backtest

One backtest in full: metrics, complete strategy settings and a deep link to the report page.

Things You Can Ask

>Which strategies beat buy-and-hold on SOL/USDT?

>Show me the top 10 backtests by Sharpe ratio and their stop-loss settings.

>How does BasicMode perform on BTC compared to simply holding BTC?

>Find backtests with over 60% return and a max drawdown under 20%.

Every answer links back to the full report page, so results stay verifiable.

More Machine-Readable Resources

OpenAPI 3.1

REST description of the read-only backtesting endpoints, if you prefer plain HTTP over MCP.

llms.txt

LLM-friendly summary of the whole site, with a full version at llms-full.txt.

MCP Server Card

Machine-readable discovery metadata under .well-known, alongside an api-catalog linkset.

Frequently Asked Questions

Is the MCP server free to use?

Yes. It is public, read-only and requires no API key, no registration and no payment.

What data does it expose?

Every completed public backtest from the unCoded Backtest Lab: return, alpha vs buy-and-hold, win rate, max drawdown, Sharpe ratio, profit factor, trade count, the full strategy settings and a deep link to the report page. Trade-level detail stays on the linked report page.

How fresh is the data?

Tool responses are served from a snapshot that refreshes about every 6 hours. Every response includes its snapshot timestamp.

Can an agent place trades through it?

No. The server is strictly read-only. It exposes historical backtest results, nothing else. It cannot trade, cannot access user accounts and is not connected to any exchange.

Are the results financial advice?

No. All results are historical backtests on specific past time windows. They are not predictions and not financial advice. Past performance does not indicate future results.

Prefer a Browser?

The same dataset powers the free Backtest Lab: leaderboards, strategy hubs and every single report page.

Open the Backtest Lab