How to get an AI research analyst running in your terminal in about fifteen minutes, and what to actually do with it once it is there. Written for investors, not engineers.
Claude Code is Claude running in your terminal. Unlike a chat window, it can read files, pull financial data, run analyses, and save its work, all by you typing what you want in plain English. For research, that means a tireless junior analyst that reads every earnings transcript, builds screens you can re-run each quarter, stress-tests a thesis, and writes up memos. The path: subscribe to Claude Max ($100/mo), paste one install line into your terminal, switch it to Opus or Fable on xhigh, then start asking. The rest of this page walks each step.
If you have only ever used the Claude chat app, this is the part worth reading slowly.
Most people meet Claude through the chat window at claude.ai. That is great for questions, but it cannot reach into your files or run anything. Claude Code is the same intelligence, living in your computer's terminal, where it can: it reads and writes documents, fetches data from the web and from data providers, runs calculations, and keeps a working folder of everything it produces.
For an investor, the mental model is simple. You have hired a sharp junior analyst who never sleeps, works only for you, and does exactly what you ask. You do not write code and you do not learn commands. You type things like "pull this company's last four calls and tell me what changed." It goes and does it, then hands you the answer and the file.
Claude Code is not on the free tier. You need a paid subscription, and the tier matters.
| Plan | Price | Good for |
|---|---|---|
| Free | $0 | The chat app only. Does not include Claude Code. |
| Pro | $20/mo | Includes Claude Code. Fine for dipping a toe in, but you will ration yourself on heavier research. |
| Max Recommended | $100/mo | The sweet spot for real research: plenty of usage so you are never rationing, plus access to the most capable models. This is the tier to start on. |
Start on Max. If you would rather test the water first, Pro works and you can upgrade in one click later. Sign up at claude.ai.
Four steps. You cannot break anything, and it is easy to remove if you ever want to.
Cmd + Space, type Terminal, and hit Enter. A plain black-and-white window opens. That is the terminal. On Windows, open PowerShell the same way from the Start menu.# macOS
curl -fsSL https://claude.ai/install.sh | bash
# Windows (PowerShell)
irm https://claude.ai/install.ps1 | iex
claude and press Enter. It opens your browser to log in. Use the same account you just subscribed with. That is it, you are in.Before you run claude, make a folder for your research so its notes land in one place. Paste these one at a time:
mkdir -p ~/Desktop/research cd ~/Desktop/research claude
There is also a Claude desktop app that runs the same thing in a normal app window. The terminal is worth the few extra minutes and is what most of this guide assumes, but the desktop app is a fine place to start if the black window feels intimidating.
One setting does more for research quality than anything else.
Claude comes in several models, and it can also think harder or faster depending on how you set it. For investing work you want maximum intelligence, not speed. Once you are in, do two things:
/model and choose Opus or Fable. These are the most capable models.That combination, a top model on xhigh, is what gives you the sharpest analysis. It is a little slower per answer, which is exactly the trade you want when you are pressure-testing an idea rather than firing off a quick lookup.
The hardest part for most people is the blank page. Here are prompts that work. Copy one, swap in your names, and go.
You do not have to get the prompt perfect. Ask, read the answer, then just say "go deeper on point three" or "now check that against the actual filing." It is a conversation, and it remembers the context.
Once the basics feel natural, this is the upgrade that makes it genuinely powerful.
Out of the box Claude can search the web, which is often enough. But the thing that turns it into a real research desk is a dedicated data feed. A good, inexpensive one is Financial Modeling Prep (FMP). For a modest monthly subscription you get an API key, a short string of characters, that lets Claude pull earnings transcripts, fundamentals, and estimates directly and reliably.
No rush on this. Get comfortable with the basics first, then add the key when you find yourself wanting cleaner data. Setting it up is a one-time, five-minute job.
A few ground rules so it makes you sharper instead of overconfident.
Once it is part of your routine, small automations compound.
After a few weeks people tend to build things they did not expect to. A running research hub that files every memo and screen in one searchable place. A short pre-market email each morning on each holding that flags anything relevant to the thesis. Screens that re-run themselves and ping you when a setup appears.
If you want to see what the finished output can look like, here is a full public research dossier built this way: coin.neilkpatel.com.
The honest pitch is this: it will not pick stocks for you, and you would not want it to. What it does is remove the grunt work between having a question and having an answer, so you spend your time on judgment instead of gathering. That is the whole game.