September 4, 2026 · Diaries
Asking Claude About My Screeners: an MCP Diary
Diary entry: I queried the Hot Prospects screener from inside Claude via MCP and it surfaced Elastic at $83.74 on a Friday. A week later it was 14.8% higher.
Last diary entry this week, and the oddest workflow of the four. No website, no API console, no alert email. Just a chat window that isn't even ours.
Friday, one prompt
StockMarketScan speaks MCP, so any AI client that supports the protocol (Claude Desktop, in my case) can query the screeners in a chat. My prompt on Friday, August 28 was one sentence: pull today's Hot Prospects list, keep only software names, and tell me which ones look like they're recovering rather than extending.
The agent called the screener tool, filtered the day's list and came back with a short table. Elastic (ESTC) at $83.74 stood out, mostly because of what the agent pointed out: ESTC had been on the same list at $85.94 on Monday the 24th. So the stock re-qualified for a momentum list while trading lower than a week earlier. Momentum criteria met on a falling price is a compression pattern I like to look at, and scrolling by hand I would have missed it.
The week after
| Date | Event | Price | Change vs Aug 28 |
|---|---|---|---|
| Mon Aug 24 | On Hot Prospects, first pass | $85.94 | +2.6% above the entry |
| Fri Aug 28 | Re-qualified lower; the MCP session | $83.74 | 0.0% |
| Mon Aug 31 | Single-session repricing, re-listed | $99.44 | +18.8% |
| Thu Sep 3 | The give-back low | $89.58 | +7.0% |
| Fri Sep 4 | Where it stands as this entry is written | $96.13 | +14.8% |
Monday did most of the work: $83.74 to $99.44 in one session. Then some of it came back. Down to $89.58 by Thursday, up to $96.13 on Friday, and the stock re-qualified for Hot Prospects on both August 31 and September 4. One week after the Friday prompt it sits at +14.8%.
Why route this through an AI at all
Fair question, the list is one click away in the browser. What I noticed doing it: the filtering happens in the same step as the lookup. "Software names that are recovering" isn't a filter the UI has, but it's easy to type. The agent also reads the whole row, which is how the lower re-qualification price came up without me looking for it. And in the same chat I can pull chart patterns and market breadth over the same MCP server and ask about all of it together. The numbers still come from the screeners. The agent just reads them faster than I do.
The honest footnote
Same disclosure as every entry in this series. This example is here because it worked. Nobody and nothing in this workflow predicted the Monday jump, and one week proves very little. The part that does repeat: one sentence turned a 415-name list into one name worth a closer look in under a minute, with the reasoning written out.
Related Resources
- MCP Server to connect the screeners to Claude, Cursor or Continue.
- Hot Prospects Screener for the list itself.
- An AI Research Workflow for a deeper MCP setup.
This article is a retrospective review of historical data for educational purposes only. It is not investment advice, the example was selected after the fact because it performed well, and screener appearances are data points, not recommendations.