wavealgo / Model Context Protocol

Your assistant writes it.
Now measure it.

Backtest Pine Script across six markets and get a shareable report, right inside your AI assistant.

Connect your assistant Try in the browser

10 free measurements / 7 days · No account or card

01 / Connect

Choose your assistant.

No wavealgo account needed

Run this command in your project’s terminal.

Terminal

claude mcp add --transport http wavealgo https://www.wavealgo.com/api/mcp

Open Claude Code and run /mcp to check that wavealgo is available.

Official guide

02 / Try it

Ask for your first measurement.

Use wavealgo to diagnose and measure a 20/50 SMA crossover strategy across all six markets. Summarize the in-sample results, fees, and limitations, and give me the report link.

Uses 1 measurement. The report link is public; ask for share: false to skip publishing a report.

Built for the research loop

Write. Check. Measure.

Fix before you run.

diagnose · free

Get compiler errors with line numbers so your assistant can fix the script before using a measurement.

Measure the idea.

measure · 1 run

Six markets, daily bars, 0.05% fee per side. See Sharpe, returns, drawdown and a report you can open.

Check the evidence.

catalog + audit · free

Look up existing public audits and their 18-cell results before testing another variation.

Server URL & other MCP clients

Use this endpoint in any client that supports remote Streamable HTTP. The free tier needs no wavealgo API key.

https://www.wavealgo.com/api/mcp

Anonymous usage is limited to 10 measurements per rolling seven days, shared by callers on the same public IP. Compiler diagnostics and catalog lookups are free.

What does the agent send?

A Pine v5 strategy body defining longSig and exitSig. The server supplies orders, fees and measurement channels. Ask the agent to call explain_harness for the full contract.

fast = ta.sma(close, 20)
slow = ta.sma(close, 50)
longSig = ta.crossover(fast, slow)
exitSig = ta.crossunder(fast, slow)

Leave out the version header, strategy() and plot(). Nested loops are rejected.

How is this different from a public audit?

MCP measurements are in-sample backtests. They do not include out-of-sample splits, repaint guards, alpha scores or audit verdicts. The public audit catalog reports those separately. Read the methodology