Build an AI stock research workflow with MCP

1. Resolve identity before analysis

Begin with symbol search when the user supplies a company name or an ambiguous ticker. Then retrieve the company profile to confirm the issuer, exchange, country, industry, currency, and whether the instrument is an ETF. This prevents a fluent analysis of the wrong security.

2. Separate reported fundamentals from market prices

Company fundamentals contains reported financial statements, ratios, and metrics. Select quarterly or annual periods and retain every fiscal period used. Price history contains normalized daily OHLCV and adjusted close; it supports returns, volatility, and event windows but not intraday execution analysis.

Do calculations inside code execution or tool composition, page by page. Return only the requested statistics, date window, method, and warnings instead of copying the raw series into model context.

3. Add event and opinion context deliberately

Earnings data provides scheduled and reported events, estimates, actual values, and surprises. Dates and estimates can change before reporting. Analyst ratings are third-party opinions, not Unquant recommendations, and should remain clearly labeled.

Processed news summaries can explain recent themes. Generated news-impact fields can help triage a story by direction, importance, and confidence, but they do not prove causality or predict returns.

4. Produce a reproducible answer

• State the resolved ticker, listing, and currency.
• Name the fiscal periods and price window used.
• Separate measured values, reported events, third-party opinions, and generated analysis.
• Retain as-of timestamps, dataset warnings, and missing-field caveats.
• Conclude with evidence and uncertainty, not a personalized buy or sell instruction.