Financial analysis is shifting from static spreadsheets to agentic workflows, as demonstrated by a recent test connecting ChatGPT and Claude to the Equibles MCP server. By running live queries on Pfizer's financials, the AI agents uncovered a stark reality: in 2025, the pharmaceutical giant paid out $9.77 billion in dividends against only $9.08 billion in free cash flow. This 108% payout ratio, derived from real-time tool calls on October 21, 2026, illustrates how large language models can now autonomously navigate complex financial statements to expose sustainability issues that might otherwise go unnoticed in a quick yield calculation.

The Agentic Workflow

The methodology relies on a four-step prompt sequence that guides the LLM through dividend history, earnings per share (EPS), free cash flow (FCF), and balance sheet debt. After connecting the Equibles MCP server—available in Claude via the Connector marketplace and in ChatGPT through the plugin store—the agent retrieves 28 ex-dates for Pfizer, revealing a stagnant quarterly dividend of $0.43 in 2025 and 2026. This pause in growth contrasts sharply with the company's financials, which show diluted EPS of $1.36 in 2025, meaning the annual dividend of $1.72 per share represents 126% of GAAP earnings.

Cash Flow Reality Check

While EPS provides a baseline, the agent's deeper dive into the cash flow statement reveals the true pressure on Pfizer's liquidity. For 2025, operating cash flow stood at $11.70 billion, but capital expenditures of $2.63 billion reduced free cash flow to $9.08 billion. Since cash dividends paid ($9.77 billion) exceeded this figure, the company had to rely on other sources to maintain the payout. The agent noted that cash dividends are not always a standard structured line item, requiring the LLM to search the 2025 10-K document directly to extract the precise figure, demonstrating the necessity of robust retrieval capabilities in financial agents.

Balance Sheet Implications

The final step of the workflow examines the balance sheet, where the agent found long-term debt rising to $61.64 billion by the end of 2025. During that year, Pfizer issued $9.68 billion in new debt while repaying only $6.76 billion, resulting in a net debt increase of $2.92 billion. This borrowing, combined with $6.93 billion spent on acquisitions, suggests that the dividend was effectively funded by leverage and asset sales rather than operational surplus. The agent's ability to correlate these disparate data points—from dividend declarations to debt issuance—provides a holistic view of financial health that simple yield metrics miss.

Key Takeaways

  • Agentic AI can autonomously execute complex multi-step financial analyses using MCP servers.
  • Pfizer's 2025 dividend payout ratio of 108% of free cash flow indicates potential sustainability risks.
  • Retrieval-augmented generation is critical for extracting non-standard financial data points like cash dividends paid.
  • The Equibles MCP server enables real-time cross-referencing of earnings, cash flow, and balance sheet data.

The Bottom Line

This case study proves that LLMs are no longer just chatbots but capable financial analysts. When paired with structured data access, they can rapidly expose the gap between headline yields and underlying cash reality, forcing investors to confront uncomfortable truths about dividend sustainability.