How to Use Claude for Stock Research: Data Sources, Prompts, Pitfalls

The short answer

Claude is a good stock research partner once it has real data to work with. There are three steps: connect the data sources your question needs, ask precise questions (and, when the stakes justify it, rigorous multi-step ones), and make Claude show its sources and dates so you can check the figures that matter. Skip the first step and Claude is working from training data and web search, which is where most bad answers come from.

Step 1: Give Claude real data

On its own, Claude answers from what it learned in training, plus whatever it finds when it searches the web. That is fine for explaining what a P/E ratio is. It is not reliable for last quarter’s margins, this morning’s guidance cut or where a stock trades against its own history. Training data goes stale. Web search can return biased, incomplete or incorrect pages, and because sources conflict, the same question can give a different answer on a different run. If you are using the answer to decide what to do with money, that is not acceptable.

Web search has three further weaknesses that matter for investing:

•           It cannot work bottom-up. An agent using web search reads the few pages that rank highest. It cannot systematically cover a universe of companies, so a screen or a pattern analysis built on search ends up resting on whatever happened to be written about and ranked.

•           It is weak on history. Search is not built around timestamps, and many articles are vague about when they were written. Asked why a stock moved on a particular past day, an agent can attach the right story to the wrong date, or the wrong story to the right one.

•           It misses information that is hidden. Useful views often sit inside podcast episodes, broker research summaries, earnings-call transcripts and filings that search does not pick up thoroughly, if it finds them at all.

Of course, it also cannot do rigorous quantitative analyses, such as historical forward P/Es or ROE, as online sources may not be available, accurate or consistent among sources.

The fix is to connect data sources through MCP (Model Context Protocol), the standard that lets Claude call outside data during a conversation. Answers then come from dated, sourced data and are the same every time you ask. A data subscription, or a free source, is a small cost next to the capital most investors are deciding on.

Data sources worth connecting

You do not need one tool that does everything. A sensible set covers a few needs:

•           Macro data: FRED. The Federal Reserve Bank of St. Louis’s FRED database is free (you request a free API key) and holds interest rates, housing, inflation, employment and thousands of other series. Several community-built FRED MCP servers exist.

•           Stock prices and volume: Yahoo Finance via yfinance. Free, and several community-built MCP servers wrap it. It also returns basic financial statements and company data. It is an unofficial route, so treat it as convenient and good enough for charts, returns and a first look at the numbers, not as a guaranteed feed.

•           Filings: SEC EDGAR. Free, and community-built EDGAR MCP servers give Claude direct access to 10-Ks, 10-Qs and 8-Ks for US companies, which is the primary source for reported numbers.

•           Narrative, qualitative data and company knowledge: Distilla. Earnings-call and filing commentary, sell-side research summaries (180K+), investor podcast summaries (11K+ episodes), company events, business drivers and explained price moves (438K+), for the most-followed companies in the US, Japan, Hong Kong, Korea and China A-shares. It brings institutional rigor to individual investors and small teams, with a free plan and paid plans from $9.90/month.

Community-built servers run on your own computer, so install only ones you trust, check that they are maintained, and keep API keys to the free, low-risk kind. Each project’s README explains how to add it.

Claude can use several connectors at once and picks between them. It helps to say which to use for what: “Use FRED for macro series, Yahoo Finance for prices, and Distilla for company news, events and research.”

Example setup: adding a remote connector

Remote servers with a web address, such as Distilla’s, are added from Claude’s settings in a couple of minutes. You need a Distilla account; the free plan includes 100 tool calls a month, with no card required.

1.         In Claude, open Customize → Connectors, click + Add, then Add custom connector. On Team and Enterprise plans, an owner adds connectors for the organization under Organization settings → Connectors.

2.         Enter Distilla as the name and https://api.distilla.ai/mcp as the URL. Leave the authentication settings as they are; Claude detects the right options.

Adding Distilla as a custom connector. The default authentication settings work as detected.

3.         Sign in with your Distilla account when the login page opens.

Sign in with your Distilla account, using Google or your email address.

4.         Click Allow to let Claude access Distilla on your behalf.

Approve the connection. You stay signed in until you sign out or revoke access.

Anthropic’s documentation notes that custom connectors are available on Free, Pro, Max, Team and Enterprise plans, with Free accounts limited to one custom connector.

Step 2: Ask precise questions, and rigorous ones

The quality of the answer tracks the precision of the question. “Is Adobe a good stock?” invites an essay. “Across Adobe’s last 12 earnings reports, was the market rewarding revenue beats or guidance?” invites analysis, because it names the company, the period and the thing to measure.

Good research questions usually pin down three things: the company or group of companies, the metric or event you care about, and the time window. Here are prompts of that kind. Copy them, then swap in your own names.

1.         “Adobe is about to report earnings. Across its last 12 prints, what did the stock do and why? Is the market rewarding revenue beats, margin beats, or guidance? Then tell me where consensus sits versus current guidance.”

2.         “Scan the events of all companies in ASIC value chain in the past 7 days and summarize key themes for each segment.”

3.         “Considering NKE as a turnaround. From its own history and peer turnarounds, what confirmed a turnaround, in what order, the lag to re-rating, and the false signals?”

4.         “Summarize NVIDIA’s price moves and reasons over the last 3 months and establish patterns.”

5.         “List every stock that moved more than 5% in the last 7 days across the US and Japan, explain each move, and group them into themes.”

6.         “Review the past earnings season for HK-listed tech stocks. Any common themes?”

Prompts 4 and 5 work because each large price move is stored with a dated cause, so Claude can look across many stocks or many months at once. That is not practical with web search.

Go further: ask for a rigorous analysis

Claude can do more than answer one-line questions. With the right data connected, it can run a multi-step analysis of the kind an analyst would build, such as a cycle study followed by an upside and downside model. The more you tell it about how to do the work, the more the result fits your own framework.

Here is an example for the US homebuilders:

“Analyze the cycle for US homebuilders and model the upside and downside for Lennar (LEN). Step 1: using FRED, pull housing starts (HOUST), building permits (PERMIT), the 30-year mortgage rate (MORTGAGE30US) and the Case-Shiller home price index (CSUSHPINSA) over the last 20 years, and identify the cycle phases. Step 2: for LEN and its closest peers, show gross margin, orders, backlog and book value per share through the last two cycles, and where each sits now. Step 3: summarize what management and the sell-side are saying about orders, incentives and margins over the past two quarters, with dates. Step 4: build bull, base and bear cases for the next 12 months with every assumption listed (mortgage rate, orders growth, gross margin, valuation multiple), and give the implied price and return for each. Step 5: tell me which assumptions the result is most sensitive to, and what would change your view. Give a source and a date for every figure.”

A few things make this kind of prompt work:

•           Spell out the framework. Say how you invest: your time horizon, whether you anchor on earnings, book value or cash flow, what you count as upside and downside, and which signals you trust. Claude will otherwise choose a generic approach.

•           Break it into steps. Cycle first, then company, then narrative, then scenarios. Each step uses different data, and you can check each one before the next builds on it.

•           Name the series and the sources. Telling Claude to use FRED for the macro series, a price source for price history and a research source for company commentary avoids guessing.

•           Make assumptions explicit. Ask for every input to a scenario to be listed, so you can change them and rerun.

•           Ask what would change the answer. A sensitivity and a list of what would prove the view wrong are more useful than a single target.

Treat the output as a first draft of your own analysis, not a conclusion.

Step 3: Check the answer

A connected Claude makes far fewer things up, but it can still misread a figure, pull the wrong period or lean on a weak source. Checking does not take long if you ask for the right things.

Ask for sources and dates. Follow up with: “For each figure, give the source and the date.” A plausible-sounding answer becomes one you can verify, and you will see quickly where the data stops.

Spot-check the numbers that matter. For a decision that rests on a single number, such as a margin or a guidance figure, open the company’s filing or results release and confirm it yourself.

Read what Claude says it could not do. Claude will often list the limits of its sourcing without being asked, such as a news window that ends on a given date or a figure it was less sure about. Those are the places to look before you act.

Run it twice. Ask the same question again, or in a new chat. If the answers differ materially, the data under them is not reliable enough to act on.

Common pitfalls

Asking without data connected. Claude will still answer, and the answer may sound confident. This is where most financial hallucination comes from. Check that the response shows a tool call, such as “Used FRED” or “Used Distilla”, before you trust any figure in it.

Using web search for screens, patterns and history. Search cannot cover a universe bottom-up, struggles with dates, and misses information inside podcasts and research summaries. Use data sources that store the information in a structured, dated way.

Vague questions. “What do you think of Shiseido?” gets you a general overview. Name the metric, the period and the comparison you want.

Relative dates. “Last quarter” can mean different things for companies with different fiscal years. Give calendar dates or name the fiscal period.

Confusing cheap with low. A stock at the bottom of its own valuation range can still be expensive in absolute terms. In fact, low multiples for cyclical stocks can mean it's close to the peak in terms of stock price and valuation.

Letting Claude pick your framework. If you do not say how you want the analysis done, you get a generic one. Tell it your approach. Inspect how the answers are biased by the sources or framework - if it's a buy/sell recommendation by a random online source or one retail investors 3 months ago, discount it. Always inspect how the analyses were done by the agent.

FAQ

Can Claude do stock research?

Yes, once it is connected to real financial data. Without a data connector, Claude answers from training data and web search, which can be out of date, biased or inconsistent from run to run.

How do I use Claude for investing research?

Connect the data sources your question needs, ask precise questions that name the company, the metric and the time window, and ask Claude for the source and date of every figure. For bigger decisions, break the analysis into steps and state your own framework. Then check the numbers your decision rests on against the original filing.

What free MCP servers can I use for stock research?

FRED (macro data, with a free API key), Yahoo Finance via community-built yfinance servers (prices, volume and basic financials) and community-built SEC EDGAR servers (filings). Community-built servers run on your own computer, so install only ones you trust. Distilla has a free plan for narrative research, company knowledge and explained price moves.

Why does Claude give different answers about the same stock?

Without a data connection, Claude relies on web search, and search results vary from run to run and conflict with each other. A connected data source returns the same dated, sourced data each time.

Can Claude screen stocks or find patterns across companies?

Not reliably with web search alone, because search cannot cover a universe of companies bottom-up. With data sources connected, Claude can scan large sets of companies, price moves and events and look for patterns.

Do I need a paid Claude plan to add a connector?

No. Anthropic’s documentation says custom connectors are available on Free, Pro, Max, Team and Enterprise plans, with Free accounts limited to one custom connector.

Does this work in ChatGPT too?

Yes. MCP is an open standard, so servers that follow it can be connected to ChatGPT as well as Claude. Check each server’s instructions. Distilla’s MCP server works with ChatGPT, Claude, Claude Code, Cursor and WorkBuddy.

Can Claude buy or sell stocks through these servers?

No. The data servers described here are read-only. Distilla never places orders.

How do I know Claude’s numbers are right?

Ask for the source and date of every figure, spot-check the ones that matter against the original filing, read any limits Claude lists, and run the question twice to see whether the answer holds.

Is this investment advice?

No. This guide describes research tools and is not investment advice. The example prompts are illustrations, not recommendations.

About Distilla

Distilla brings institutional rigor to individual investors and small teams. It is the fundamental research layer for AI agents: why stocks move, not just where. The same data powers an enterprise product, including an agent, used by some of the world’s top funds, and is also packaged as a self-serve product so individuals can work from the same facts. It is an MCP server and HTTP API that gives Claude, ChatGPT and your own agents quantitative data, qualitative research and structured company knowledge in one queryable, read-only layer.

•           Quantitative: financial statements, analyst consensus, valuation history and company metrics, including AI-extracted segments and industry-specific KPIs.

•           Qualitative: earnings-call and filing commentary, 180K+ sell-side research summaries and 11K+ investor podcast episodes.

•           Company intelligence: 1,000+ company knowledge topics, business drivers updated quarterly, company events (about 50,000 added weekly), products, peers and sector relationships, and 438K+ explained price moves of 3% or more.

Core coverage is the most-followed companies in the US, Japan, Hong Kong, Korea and China A-shares; sell-side and podcast summaries extend beyond the core universe. Read-only MCP server at https://api.distilla.ai/mcp, with the same lookups over an HTTP API. Works with Claude, ChatGPT, Claude Code, Cursor, WorkBuddy and Doubao. Self-serve: free plan, with paid plans from $9.90/month. Enterprise: contact Distilla. Distilla is a research tool, not investment advice. Figures are as of 7 October 2026.