The screener speaks in filters: P/E below 25, ROCE above 15, volume ratio above 2. Claude speaks your language and translates. That makes the connected version the fastest way to go from a loose idea to a ranked shortlist.
You need a connected AI client first.
Describe the idea, not the filters
"Find me quality midcaps that look undervalued and have shown unusual volume this week"
Claude maps that to real criteria: "quality" becomes ROE and ROCE thresholds with a debt ceiling, "undervalued" becomes P/E or P/B limits, "midcap" picks the right universe, and "unusual volume" attaches a volume spike signal. Then it runs the screen and reports:
- What it screened for, so you can sanity-check the translation
- A results table with the metrics that matter for your criteria
- Standouts: two or three names that look particularly interesting and why
- Refinement ideas if the list came back too long or too short
Filters combine with AND logic (every filter must pass) while signals combine with OR logic (any firing signal qualifies), which is the same engine the website screener uses. You can mix both in one request.
The refinement loop
The first screen is rarely the right screen. Iterate in plain language:
- "Too many results. Add a condition that revenue grew in the last year."
- "Drop anything below 1,000 Cr market cap."
- "Now sort by how far they are from their 52-week high."
- "Rerun the same screen but on the Nifty 500 only."
Because it is one conversation, each refinement takes seconds and the assistant remembers the intent behind the screen, not just the filters.
Scanning for signals
Screens filter on levels; scans catch events. Ask directly:
- "What stocks are breaking out to 52-week highs today?"
- "Show me large caps with volume spikes right now."
- "Any oversold bounces setting up in the pharma sector?"
There are 15+ signal types available, from breakouts and gap moves to moving average crosses. During market hours a subset updates live; the rest are computed end-of-day, and Claude will tell you which is which.
From shortlist to decision
The best part of screening in a conversation is that the next step is zero-friction:
Screen
Run the idea, refine until the list is 5 to 15 names.
Triage
"Get the conviction score for all of these." Names where multiple independent signals converge earn the next 15 minutes; the rest can wait.
Research the survivors
Pick the top candidate and say "Research it." You are now in the end-to-end research workflow without switching tools.
Track the maybes
"Add the top three to my watchlist and alert me if any of them spikes on volume." The screen keeps working for you after the conversation ends.
A screen produces candidates, never conclusions. Whatever comes back still needs the research pass, and nothing in the results is a recommendation to buy or sell.
Connect your AI client