# How to research stock sentiment with an AI assistant > A step-by-step tutorial for researching stock sentiment inside Claude or ChatGPT: connect Quantral's MCP server, ask what is trending, read a score with its tier, and catch up on months of chatter in one question. By Maya Koeva · 2026-09-04  You can now ask an AI assistant what investors are saying about a stock and get an answer built from measured data instead of the assistant's memory. Quantral runs an MCP server: connect it to Claude or ChatGPT once, and the assistant can pull the same scored, credibility-weighted [sentiment](/learn/what-is-market-sentiment) readings the Quantral app shows, in the middle of whatever conversation you are already having. This tutorial walks through the setup and the questions that do most of the work. If you want the app-side version instead, **[the Quantral tutorial](/learn/how-to-use-quantral)** covers that in five steps. ## Why route this through an assistant at all Ask a plain chatbot "what's the sentiment on SanDisk?" and it answers from training data that is months old, or from a quick web skim of whoever happens to rank. It has no way to tell a loud account from a right one. Connected to Quantral, the assistant stops guessing. It queries the conversation Quantral tracks across finance X, Reddit, and Substack, where each stock carries a 0 to 100 [signal score](/learn/what-is-a-stock-signal) and each author carries a graded [credibility score](/learn/what-is-a-credibility-score) based on how their past calls resolved. The assistant's job becomes phrasing and follow-up; the numbers come from the data. And because it is a conversation, you can push on anything: compare two names, ask what changed since July, ask who is behind a reading. ## Step 1: Connect Quantral to your assistant Open [app.quantral.com/connect](https://app.quantral.com/connect) and follow the steps for your client. Claude and ChatGPT both work, as does any MCP-compatible agent; the server lives at `https://app.quantral.com/api/mcp`. You sign in with your Quantral account and approve read access. MCP access is included in every plan, and new accounts start with a 7-day free trial. Two things worth knowing before you approve. The connection is read-only: the assistant can fetch signals, scores, and recaps, and it cannot touch your watchlists, account, or billing. And in Claude and ChatGPT the responses render as interactive cards in the chat, so a ranked list arrives as a list you can scan, not a wall of JSON. The **[/mcp page](/mcp)** has the full rundown of what the server exposes. ## Step 2: Ask what is trending Start wide. The prompt: What names are trending in the last 24 hours? Asked on the morning of September 4, 2026, that returned a ranked card of the five strongest signals across the accounts Quantral tracks: Ciena at 89 (Strong), Nebius at 88, NVIDIA at 86, SpaceX at 86, and Robinhood at 83, each with a one-line reason. The Ciena line is a good example of why the reason matters: a record quarter, and the stock down 10% anyway. A score with the reason attached hands you a research question worth running down. Each name in the card links back to its Quantral page, so the jump from "interesting" to "show me every call behind this" is one click. ## Step 3: Read one stock's score Then go narrow: What is the signal score for SNDK? Same morning, the answer: SanDisk at 81 over the last 24 hours and 83 over 7 days, both in the Strong tier, with the current conversation centered on a NAND demand forecast doing the rounds on X. Every score comes back with a tier word attached: Exceptional, Strong, Moderate, Weak, or Negative. The two windows matter more than either number alone. A high 7-day score with a sagging 24-hour score is a conversation cooling off; the reverse is one heating up. If a stock comes back with no score, that means the accounts Quantral tracks produced no mentions of it in the window. Treat that as a coverage boundary, not as proof that the market has no opinion. ## Step 4: Catch up on a name in one question The recap tool is the one that saves the most time: Summarize last month's chatter recaps for AMD. For August 2026 that returned a Moderate reading of 72, up 20 points on the month, built from 230 posts across Reddit and X: an earnings beat on record revenue, an AI chip acquisition the crowd mostly cheered, and a criticism that surfaced later in the month, compressed into a paragraph with the main authors named. You can page further back month by month, which turns "I have not looked at this stock since spring" into a two-minute catch-up. ## Step 5: Push on the reading The setup earns its keep on follow-ups, because the assistant can chain questions you would otherwise answer by hand: - "Compare the sentiment on AMD and INTC this week." Two scores, two tiers, one answer. - "Who is behind the SanDisk reading?" The individual calls, with each author's graded record. - "Has the NVDA conversation changed since July?" Monthly recaps, read side by side. The habit that makes this useful is the same one the app teaches: treat the score as a reading of the conversation, then check who is doing the talking before you weight it. ## What it will not do The score measures how strong and credible the conversation around a stock is. It is not a price prediction, and the assistant will tell you the same thing, because every response carries that framing. Quantral reads a curated set of accounts, so a quiet reading describes our coverage of a name, not the whole market. And the connection stays read-only end to end; nothing you ask can change your account. Answers above are snapshots from the morning of September 4, 2026. For the current reading on any name, ask your assistant, or open the [Quantral app](https://app.quantral.com) and see the score, every call behind it, and the track record of each account making them.