How to Track What Hedge Funds Are Buying — Directly From Your AI Agent

The $45 trillion data set that most investors never see — and how AI just made it accessible to everyone

Imagine sitting at your desk on a Monday morning. Your AI assistant has just finished scanning the latest SEC filings. It tells you: "Three top hedge funds initiated new positions in the same cybersecurity stock this quarter. The CFO bought shares on the open market last week. And the stock hasn't moved yet."

That is not a Bloomberg Terminal. That is not a $20,000-a-year data subscription. That is what happens when you connect a freely available financial dataset to an AI agent that knows how to read it.

For decades, one of the most powerful sources of investment intelligence has been sitting in plain sight — buried in text files on a government website, readable only by machines and the handful of Wall Street analysts who knew how to parse them. That data set is called 13F filings, and it represents over $45 trillion in institutional holdings, updated every single quarter.

The catch has always been accessibility. Until now.

The Best-Kept Secret on Wall Street

Here is something most people do not realize: under Section 13(f) of the Securities Exchange Act of 1934, every institutional investment manager with over $100 million in assets must file a quarterly report disclosing every single one of their U.S. stock holdings. Not a summary. Not a highlight reel. The full list.

That includes:

Hedge funds like Bridgewater, Citadel, and Renaissance Technologies
Mutual fund giants like Fidelity, Vanguard, and BlackRock
Pension funds managing billions for teachers, firefighters, and public employees
Family offices of billionaires you read about in the news
Insurance companies and trust banks with century-long investment horizons

Over 5,000 institutions file every quarter. The combined value of their disclosed holdings exceeds $45 trillion. And thanks to a law passed during the Great Depression, it is all public, free, and legally mandated.

So why do 99% of retail investors never look at this data?

Because raw 13F filings are a nightmare to work with. They arrive as text files on the SEC's EDGAR system — thousands of lines of cryptic CUSIP identifiers, share counts, and dollar values, with no company names, no ticker symbols, no charts, and certainly no analysis. Turning that into something a human can read requires serious data engineering.

Or, as of very recently, a single API call.

Enter MCP: The Protocol That Lets AI Agents Talk to the Real World

If you have been following the AI space this year, you have probably heard about the Model Context Protocol (MCP). It is an open standard — created by Anthropic and now supported across the industry — that allows AI agents to interact with external tools and APIs in a standardized way.

Think of MCP as the USB-C of AI integrations. Before MCP, every connection between an AI agent and an external data source was a custom integration — fragile, one-off, and expensive to maintain. With MCP, you plug in a server once, and any MCP-compatible agent (Claude, ChatGPT, Cursor, and increasingly everything else) can use it immediately.

Most MCP servers today connect to developer tools: file systems, databases, GitHub repositories, documentation sites. That makes sense — developers were the first to adopt AI agents, and they built the tools they needed.

But there is an entire world of data outside of localhost:5432.

What if your AI agent could answer questions like:

"Which stocks are the top hedge funds buying right now?"
"Show me every position Berkshire Hathaway added to last quarter."
"Are company executives buying or selling their own stock?"
"Find tickers where both institutions AND insiders are accumulating."

That is exactly what happens when you connect a financial data MCP server to your AI agent.

The Seven Tools Your AI Agent Needs

A few months ago, I stumbled across a free, open-source MCP server that processes 13F filings and insider trading data into clean, queryable formats. It exposes exactly seven tools — each one mapping to a specific real-world question an investor might ask.

Here is what it gives your AI agent:

1. Search Institutions — "Find me every institution with 'Berkshire' in its name." Returns the institution's slug, AUM, and behavioral scores.

2. Institution Profile — "What is Bridgewater's investment style?" Shows concentration level, turnover rate, sector conviction — essentially a personality profile for any fund.

3. Institution Holdings — "What does Michael Burry's fund actually own?" The full 13F holdings list — ticker, market value, portfolio weight, and whether they added or reduced last quarter.

4. Stock Overview — "Who owns NVDA, and is institutional money flowing in or out?" Holder count, net institutional flow, top holders ranked by conviction and by dollar value, plus ownership history over time.

5. Stock Flows — "Rank every stock by net institutional buying this quarter." A ranked leaderboard of where smart money is moving — in either direction.

6. Insider Activity — "Is NVIDIA's CEO buying or selling?" Form 4 data showing every trade by officers, directors, and 10% owners, plus a confidence score per ticker.

7. Smart Money Convergence — the flagship signal. This is where it gets really interesting. This tool cross-references the two completely independent data sources — 13F institutional filings and Form 4 insider trades — and surfaces tickers where both groups are buying at the same time. When the people managing billions and the people running the company are both accumulating, that is not noise. That is a signal.

What Makes This Different

If you search for "financial data API," you will find dozens of them. Most charge hundreds or thousands of dollars per month. They are built for hedge funds and trading desks, not for curious individuals experimenting with AI agents.

If you search for "MCP servers," you will find hundreds of them too. Almost all are built for developer infrastructure — databases, cloud providers, version control. None of them answer the question "what is smart money doing right now?"

This one sits at the intersection: financial intelligence delivered through the protocol that AI agents already speak.

And the best part? It works without an API key, without a credit card, without even creating an account. You can run it right now:

npx alphasmo convergence --limit 5

npx alphasmo convergence --limit 5

That single command queries the latest 13F data, computes the convergence signal, and prints a table of the top tickers where institutional and insider buying are both flashing green. No setup. No database. No data ingestion pipeline. Everything is handled by the platform and served through the API that the MCP server wraps.

Under the hood, the data pipeline is non-trivial: CUSIP resolution (mapping those cryptic 9-character identifiers to actual ticker symbols), quarterly filing deduplication, flow computation against prior quarters, personality scoring based on historical behavior patterns — all the messy stuff that makes raw SEC data so difficult to work with. The MCP server abstracts all of that away behind seven clean function calls.

How to Wire It Up

If you are using Claude Code (the terminal-native AI coding agent), it is literally one command:

claude mcp add alphasmo -- npx -y alphasmo@latest mcp

claude mcp add alphasmo -- npx -y alphasmo@latest mcp

For Claude Desktop, Cursor, or any other MCP-compatible client, add this to your configuration file:

{
  "mcpServers": {
    "alphasmo": {
      "command": "npx",
      "args": ["alphasmo", "mcp"]
    }
  }
}

{
  "mcpServers": {
    "alphasmo": {
      "command": "npx",
      "args": ["alphasmo", "mcp"]
    }
  }
}

That is it. Your AI agent now has access to:

5,000+ institutional investors with detailed behavioral profiles
$45 trillion in quarterly holdings data, searchable by ticker or institution
Form 4 insider trading with confidence scoring
The smart money convergence signal — unique to this data set

You can optionally add a free API key (grab one from the developer dashboard in 30 seconds) to raise the anonymous rate limit, but it is not required to start experimenting.

What You Can Actually Do With This

Let me give you a few concrete workflows. These are not hypothetical — I have run every one of them.

Morning Market Scan

Ask your agent: "What are the top 10 stocks institutions are buying this quarter, and are insiders confirming the signal?"

The agent calls get_stock_flows with direction=buy, then cross-references each ticker with get_insider_activity. Within seconds, you have a list of stocks where both signals align — without opening a single website or spreadsheet.

Deep Dive on One Institution

Ask: "Show me Renaissance Technologies' latest 13F — what did they add, what did they cut, and how concentrated is their book?"

The agent calls get_institution_holdings for the full position list and get_institution_profile for the personality scores. You learn not just what they own, but how they invest — are they making concentrated bets or spreading across hundreds of names?

Insider Watch

Ask: "Alert me to any ticker where three or more insiders bought shares this week."

The agent calls get_insider_activity for stocks you care about. Form 4 filings are due within two business days of the trade — compared to 13F's 45-day lag, this is near-real-time intelligence.

Convergence Screening

Ask: "Show me tickers where institutional buying AND insider buying are both in the top quartile this quarter."

This is the flagship get_smart_money_convergence tool. It combines the two independent data sources — 13F and Form 4 — into a single scored signal. When independent, legally-mandated data sources agree, the signal-to-noise ratio goes up dramatically.

The Bigger Picture

We are in the middle of a fundamental shift in how people access and analyze information. AI agents are not just chatbots — they are becoming the interface layer between humans and the world's data. And MCP is the protocol that makes that interface universal.

Most of the world's valuable data is not sitting in a neatly documented REST API. It is buried in government filing systems, regulatory databases, and public records that were designed for compliance, not for consumption. The gap between "this data exists" and "this data is useful" has always been enormous — and it has always been bridged by expensive, proprietary products.

What MCP changes is the economics of that bridge. When the interface is standardized, the cost of connecting a data source to an AI agent approaches zero. And when the cost approaches zero, entirely new use cases become viable.

A retail investor asking their AI assistant "what are the pros buying?" is not a Bloomberg Terminal replacement. It is something different — and arguably more important. It is the democratization of financial intelligence that has been locked behind paywalls and data engineering complexity for decades.

The 13F data has been public since 1978. The tools to make it accessible to ordinary people have only existed for a few months.

One Command Away

If you have an MCP-compatible AI agent set up, you are one command away from giving it access to the same institutional holdings data that professional analysts use:

npx alphasmo convergence --limit 10

npx alphasmo convergence --limit 10

No signup. No credit card. No setup. Just real financial data, delivered through the protocol your AI agent already speaks.

The $45 trillion question is: what will you ask it?

— -

Disclaimer: This article is for informational and educational purposes only. It does not constitute investment advice, solicitation, or recommendation. 13F filings have a 45-day reporting lag and may not reflect current holdings. Form 4 data reflects historical transactions. Past performance and institutional behavior do not guarantee future results. Always conduct your own due diligence before making investment decisions. The tools and services mentioned are provided by third parties; no endorsement or guarantee of accuracy is implied.

Contents