• Why clusters, not single buys
• What it returns
• A real worked example
• The bigger picture
• Installing in 60 seconds
• Try it

Insider buying is one of the oldest legal signals in markets and one of the worst-organized. The SEC files every transaction at EDGAR. Bloomberg charges institutions for clean access. Retail gets a tangled mess of XML, late filings, and 50-row tables with no context.

For an AI coding agent — Claude Code, Cursor, Cline, Windsurf — that gap is doubly painful. The model is fluent enough to reason about Form 4 patterns. But without a clean data feed, it can only write parsing code that doesn’t run, or refer you to OpenInsider screenshots from 2019.

This post shows how the insider_buy_clusters tool inside the FalsifyLab Alpha MCP server closes that gap, and why insider clusters specifically are the signal that survives the most academic scrutiny.

§Why clusters, not single buys

A single 10-K Form 4 filing from one director buying $200K of his own stock is noise. The full insider trading literature (Jeng-Metrick-Zeckhauser 2003, Cohen-Malloy-Pomorski 2012) found the alpha is concentrated in two patterns:

• Clusters — three or more distinct insiders at the same company buying within a narrow window (typically 5-30 trading days)

• Opportunistic vs routine — insiders who don’t trade on a schedule, buying right before a fundamental event

The cluster signal is robust across decades. The single-buy signal mostly isn’t.

insider_buy_clusters is built around that finding. It scans EDGAR’s daily Form 4 feed, groups by ticker, counts distinct filers, deduplicates 10-K vs 10-Q context, and returns clusters that meet a configurable threshold (default: 3+ distinct insiders within 10 trading days).

§What it returns

A typical call:

insider_buy_clusters(lookback_days=30, min_insiders=3, sector=null)Returns:

• ticker

• count of distinct insider filers

• total dollar value

• earliest and latest filing dates in the cluster

• list of (name, title, shares, dollars) per filer

• post-cluster price reaction (1-day, 5-day, 30-day)

• filing-link list to original EDGAR XML for audit

An AI agent can then ask “of these 14 clusters, which ones had CEO + CFO in the same cluster” or “filter to clusters where total dollars > $5M and the company hasn’t had any insider buying in the prior 12 months” — both are one-line follow-up tool calls.

§A real worked example

User asks Cursor: “Show me the strongest insider buy clusters in semiconductors in the last 60 days, and tell me which ones happened before significant share-price moves.”

Without MCP: the agent writes 200 lines of EDGAR XML scraping code that times out.

With FalsifyLab Alpha MCP, the agent:

• Calls insider_buy_clusters(lookback_days=60, sector="semiconductors", min_insiders=3)

• Receives 4 cluster events with ticker, filer list, total dollars, 30-day price reaction

• Calls confluence_today to cross-reference whether ETF flows or 8-K filings aligned

• Synthesizes a 3-paragraph answer naming specific tickers and dates

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

§The bigger picture

insider_buy_clusters is one of 13 tools in the FalsifyLab Alpha MCP server. The differentiator isn’t any single tool — it’s the cross-source confluence layer that stacks insider clusters with 8-K filings, ETF flows, on-chain wallets, DeFi yields, and Polymarket whales into a single “where do these signals align” call.

That’s confluence_today. No other MCP server in the space stacks signals across data sources like this.

§Installing in 60 seconds

pip install falsifylab-alpha-mcpThen add to your agent’s MCP config. For Claude Code:

{ "mcpServers": { "falsifylab": { "command": "falsifylab-alpha-mcp", "args": [] } } }Cursor, Cline, Windsurf use the same shape. Restart the agent, see 13 new tools (3 free + 10 Pro) in its catalog.

Free tier has 3 tools (top_yield_farms, hl_vault_leaderboard, macro_tape) with no signup. Pro at $19/month unlocks insider_buy_clusters plus 9 other live tools.

§Try it

If you’re using Claude Code or Cursor for any equity or trading research, the free install is genuinely free. Try the three free tools first to confirm the integration works end-to-end. If insider_buy_clusters or confluence_today save you 20 minutes a week, Pro pays for itself in week one.

falsifylab.com · insider_buy_clusters guide · confluence_today guide · pricing

— research and educational content. not investment, legal, or tax advice. do your own research. positions and views may change without notice.

Originally published on FalsifyLab Substack.