AI Agents for Financial Filings: MCP, Claude and Automated Disclosure Screening

Learn how to use AI agents for financial filings by connecting language models to live data using the Model Context Protocol and automated APIs.

Language models are good at reading filings and bad at knowing what happened yesterday. The fix is to connect them to live, structured data. There are two practical ways to do it for financial disclosures: give an assistant tools through the Model Context Protocol (MCP), or call an agent API that does the reading and checking for you.

Key takeaways

• MCP lets Claude, ChatGPT, Cursor and other clients call external tools with live data instead of guessing from training data.
• The Bourse MCP server is listed in the official MCP Registry and exposes datasets, ticker profiles, insider clusters and Congress and Trump trades.
• For automated pipelines, an agent API returns a fixed JSON decision for each filing, which is easier to plug into alerts than free-form chat.
• Ground every "political link" in real records. Models invent connections when they are not given data.

Option 1: MCP for assistants

MCP is an open standard for connecting AI clients to tools. You add a server URL and a key; the assistant discovers the tools and calls them when your question needs them.

`` URL: https://api.bourse.pl/mcp Header: Authorization: Bearer bk_... ``

Then ask in plain language:

• "Which senators bought defense stocks in the last 60 days?"
• "Show insider buying clusters this month and which of those companies also won federal contracts."
• "Did any German board members buy shares of their own company this week?"

The assistant answers from Bourse data and can show the rows it used. Bourse is listed in the official MCP Registry as io.github.ofylypczuk/bourse, and the code is on GitHub.

Option 2: an agent API for pipelines

Chat is great for exploration and awkward for automation. If you screen hundreds of filings a day, you want the same JSON every time. Bourse Agent does exactly that:

``bash curl -X POST https://api.bourse.pl/v1/agent/analyze \ -H "Authorization: Bearer $BOURSE_KEY" \ -d '{"url": "https://.../current-report.pdf"}' ``

It reads the document in any language (ESPI and EBI reports, BaFin and AMF notices, RNS announcements, SEC 8-K and Form 4), checks it against insider trades in the U.S. and Europe, Congress trades and committee seats, lobbying and federal contracts, and returns:

``json { "verdict": "noise", "tone": "neutral", "confidence": 0.82, "next_step": "Archive: routine option exercise, no links to defense committees.", "political_links": [], "insider_activity": "No other director dealings in 12 months" } ``

Your pipeline can route "material" to analysts, "watch" to a daily digest and drop "noise".

Avoiding hallucinated connections

The biggest risk with AI in compliance and research is invented facts, especially links between politicians and companies. Two rules help:

1. Give the model the data. The agent fetches Congress trades, committee seats, lobbying and contracts for the issuer before it judges.
2. Constrain the output. Bourse Agent may only list political links that exist in that data; if there are none, the list is empty.

Limits and pricing

The free key includes 10 agent calls a day and full MCP access within normal API limits. API Pro raises the agent to 2,000 calls a day and allows commercial use. Pay-per-call through AWS Marketplace and a Docker image for your own AWS account are coming; see integrations.

Summary

Use MCP when people ask questions, and an agent API when software makes decisions. Both work best when the model is grounded in timestamped public filings rather than its memory.

Bourse provides financial data and editorial information, parts of which are generated by AI. This page is for general education only. It is not personalised investment advice or a recommendation to buy or sell any financial instrument. Make your own decisions.