• What the gap actually feels like
• Why FalsifyLab specifically
• The 13 tools
• Installing
• A worked example
• What this is not

If you’ve tried using an AI coding agent for trading research lately, you’ve probably hit the same wall everyone else has. The model can write beautiful pandas code. It can reason about your trading idea. It can debug your backtest. But ask it “did anyone insider-buy NVDA last week” and it shrugs — its training data ends months ago.

That gap is exactly what MCP (Model Context Protocol) was built to close.

This post walks through the live finance data layer we built for AI coding agents, called FalsifyLab Alpha MCP. 13 tools. 3 free. 10 Pro at $19/month. Stdio MCP server, Python stdlib only, zero runtime deps.

§What the gap actually feels like

A typical session before MCP existed:

You: “Pull up the most recent insider buying clusters for semiconductors.” Agent: “I’d need real-time SEC data. I can show you how to query EDGAR yourself.” You: “Just do it.” Agent: writes 80 lines of XML parsing code that doesn’t run

After MCP:

You: same question. Agent: calls insider_buy_clusters tool with sector=semiconductors, days=14 → gets back 6 cluster events with Form 4 IDs, exec names, dollar amounts, post-buy returns. Agent: summarizes the 3 that stand out.

The agent isn’t smarter. It just has a reliable data tap.

§Why FalsifyLab specifically

DeFiLlama has a great API for protocol TVL. Alpha Vantage has equity prices. EDGAR has filings. Bloomberg has institutional data if you have $25K/year. Each of these is a single slice.

What was missing: a unified MCP server spanning SEC filings + ETF flows + on-chain wallets + DeFi yields + Polymarket + Hyperliquid vaults — with a cross-source confluence layer that flags tickers where two or more signals align in the same 24-hour window.

That confluence layer is the differentiator. confluence_today is one tool call that surfaces “AAPL had insider buys AND ETF inflows AND post-earnings drift in the last day.” No other MCP server stacks signals across sources like this.

§The 13 tools

Free tier (no signup required):

• top_yield_farms — DeFi APYs with token emissions stripped and impermanent-loss risk priced

• hl_vault_leaderboard — Hyperliquid public vault performance ranking

• macro_tape — DXY, gold, oil, 10-year yield, MOVE index in one call

Pro tier ($19/month, real-time, cache bypassed):

• insider_buy_clusters — SEC Form 4 clusters with executive context

• sec8k_material_today — Item 1.01, 2.01, 5.02 material 8-K filings

• etf_flow_today — US spot BTC and ETH ETFs, daily net flows

• polymarket_whale_positions — open positions over $10K across all markets

• onchain_smart_wallets — Solana wallets scored by realized PnL

• active_airdrop_farms — eligibility checklists for current airdrops

• confluence_today — the cross-source signal alignment tool

• earnings_drift_radar — post-earnings surprise plus reaction direction

• token_unlock_radar — upcoming token unlocks by date and notional

• fed_comm_radar — Fed speaker calendar tagged with hawkish/dovish tone

§Installing

pip install falsifylab-alpha-mcpThen point your agent at it. Claude Code config (~/.claude/mcp_servers.json):

{ "mcpServers": { "falsifylab": { "command": "falsifylab-alpha-mcp", "args": [] } } }Cursor, Cline, Windsurf use the same mcpServers shape under their respective config paths. Restart the agent, see the new tools appear in its catalog.

§A worked example

You: “Did anyone insider-buy NVDA in the last 30 days, and did any ETF accumulate aggressively at the same time?”

The model:

• Calls insider_buy_clusters with ticker=NVDA, lookback_days=30 — receives Form 4 cluster events with names, dates, dollar amounts.

• Calls etf_flow_today — checks if any equity ETF holding NVDA had concentrated inflows in the same window.

• Calls confluence_today — sees whether the system already flagged NVDA cross-source.

• Synthesizes a 3-paragraph answer with specific cluster dates and ETF flow dollars.

Total round-trip: under 4 seconds. Total user effort: typing the question.

§What this is not

Not investment advice. Not a signal service. Not a hedge fund. FalsifyLab Alpha is a research data layer with a transparent pricing tier. The same data feeds a public trading bot fleet whose results we publish honestly, including the bots that died.

If you’re using Claude Code or Cursor for any serious trading or financial research, the install is 60 seconds and the free tier is genuinely free. Try the three free tools first. If confluence_today or insider_buy_clusters save you 20 minutes a week, Pro pays for itself.

falsifylab.com · install guide · browse all 13 tools · pricing

Originally published on FalsifyLab Substack.