How to Connect Claude Code to Live Stock Market Data (One Command)

TLDR

• Claude Code talks to live market data through MCP (Model Context Protocol). One claude mcp add command connects the SNACS server, and from then on you ask questions in plain English and get answers backed by the scanner, dilution facilities, and SEC filings.
• Once connected, Claude Code can pull real numbers: GCDT's +224.9% TRUE MFE on 410.1M shares (Sep 8), IMRN trading 4,382.1x its average daily volume (Sep 4), and DSS running from $0.60 to $1.18 the same session it filed to pull a public offering.
• The value is not the chat window. It's that Claude Code can cross-reference a volume spike against the ticker's catalyst, cash runway, and active dilution facilities in a single prompt — the exact join a human does by hand across five tabs.
• Worked examples below use GCDT (Industrials, +133.1% last week) and IMRN (Pharmaceuticals) to show the difference between a query that returns noise and one that returns a tradeable read.
• This is the same live data behind the SNACS scanner and SEC research tools — MCP just moves it into your terminal.

What Connecting Claude Code to Live Market Data Actually Means

Connecting Claude Code to live market data means giving the model a direct, structured pipe to real-time scanner and filing data through MCP, so it answers from today's tape instead of its training cutoff. MCP is an open protocol that lets an AI client call external tools and data sources in a standard format. Claude Code is the client; the SNACS MCP server is the tool provider. When you connect them, Claude Code gains a set of callable functions — query the scanner, pull a ticker's dilution snapshot, list recent SEC filings — and decides on its own which to call based on your question.

The distinction that matters: a base language model answers stock questions from memory, and its memory is frozen and frequently wrong on small-caps. Ask a disconnected model about a $0.40 industrial name and it either hallucinates or refuses. Ask Claude Code with the MCP server attached and it calls a live endpoint, gets back the actual session data — premarket, regular-hours, after-hours OHLC, volume, RVOL, float, cash runway — and reasons over that.

This is not a chatbot bolted onto a price feed. The reason MCP is worth setting up is composition. A single prompt like "which stocks last week ran over 100% and what were their catalysts" forces the model to chain calls: hit the runner list, then for each name pull the catalyst and the filing history, then synthesize. A human does that join across the scanner, a news tab, and the SEC filing browser. Claude Code does it in one turn because each of those data sources is exposed as an MCP tool it can call in sequence.

The data Claude Code can reach through the SNACS server is the same universe streaming in the scanner: 2,500+ tickers with 30+ columns each, plus the SEC filing corpus (1.4M+ filings) and the dilution facility database. That last piece is what separates a real market-data connection from a toy. The model can ask "does this ticker have an active shelf or ATM" and get back facility counts — the difference between a clean momentum name and a stock about to sell into your buy order.

If you want the deeper engineering context on how the server is built and what it exposes, we wrote that up separately: Building an MCP Server for Live Market Data: What We Learned. This article is the trader-facing version — get it connected, then use it.

The One Command

Connecting takes a single claude mcp add call from your terminal — Claude Code handles the handshake and the tools appear on the next prompt. From a shell where the claude CLI is installed:

claude mcp add snacs --transport http https://mcp.snacs.trade

That registers the SNACS server under the name snacs. Restart your Claude Code session (or start a new one) and the market-data tools are live. You can confirm with:

claude mcp list

which prints the registered servers and their connection status. If snacs shows connected, Claude Code can now call it.

That's the whole setup. No API-key juggling in a config file you'll forget about, no local process to babysit — the HTTP transport means the server runs remotely and Claude Code reaches it on demand. If you use other editors, the same server connects to them too; we've covered VS Code Copilot, Cursor, and the Claude web and desktop apps in their own guides. The protocol is identical; only the add syntax differs per client.

Now the part that actually matters — what to ask it.

Worked Example #1: Asking Claude Code to Explain GCDT's +224.9% MFE Day

GCDT is the cleanest recent case of a query that returns a full read, because the volume, the move, and the catalyst all landed in the data on the same day. Last week (Aug 31–Sep 4) GCDT was the top runner at +133.1%, moving $0.40 to $0.94 on 410,132,793 shares in the Industrials sector. The single biggest session came Sep 8: regular-hours open $0.36, high $1.17, low $0.36, close $0.94, with an after-hours close of $0.71. Full-day TRUE MFE (low to high across all sessions) was +224.9% on 410.1M shares.

A disconnected model can't tell you any of that. Connected, the prompt "pull GCDT's session data and recent filings, and tell me what drove the Sep 8 move" makes Claude Code call the scanner for OHLC, then the filing endpoint. It comes back with the price action above plus the catalyst: GCDT filed a 6-K on Sep 8 and announced a strategic partnership for mass production of BocaPCM-TES panels in China. That's the driver, stated as a filing, not a guess.

Why the profit math is worth running through: use a $10,000 base position. Capturing the full MFE from the $0.36 low to the $1.17 high would have returned $32,490 (+224.9%). Nobody nails the exact tick, so the honest number is the range — the open-to-high leg from $0.36 was the same move, and even a partial capture off the open ran multiples. The point of asking Claude Code isn't to fantasize about the top tick; it's that within seconds you have the low, the high, the close, the volume, and the filing that triggered it, and you can decide whether the next GCDT-shaped setup is worth your size.

Contrast the after-hours fade: GCDT closed the regular session at $0.94 but the after-hours close was $0.71. If your only data point was "it closed at $0.94" you'd misjudge overnight risk. The MCP server exposes all three sessions, so Claude Code sees the AH give-back and can flag it. That session-awareness is the difference between a data connection and a headline.

Worked Example #2: Cross-Referencing IMRN's Volume Spike With Its Catalyst

IMRN shows why the join — not any single number — is the reason to connect the model. On Sep 4, IMRN traded 123.1M shares, 4,382.1x its 50-day average daily volume. Premarket high $1.98; regular hours opened $1.81, high $1.99, low $1.54, closed $1.77. Full-day range $1.16–$1.99, TRUE MFE +71.5%, but the regular-session close was -2.2%. Over the last four sessions IMRN ran $1.08 to $1.82 (+69.0%) on max daily volume of 123,096,607.

Here's the read a connected model surfaces in one turn. RVOL of 4,382.1x is not normal accumulation — that's a catalyst-driven volume regime change. Claude Code, prompted with "why did IMRN spike on Sep 4," pulls the catalyst: Immuron announced it will launch PROIBS® in the United States (Sep 4). Then the useful part — it also pulls the cash tier and dilution facilities, because a 4,000x-RVOL day on a micro-float pharma name is exactly when an offering gets shoved out the door. That single-prompt chain (spike → catalyst → dilution check) is the entire argument for MCP over a static price lookup.

Compare the two examples side by side and the lesson sharpens:

Both had a real catalyst. GCDT closed green and held most of the move; IMRN closed red on the session despite a +71.5% intraday MFE. Same tool, two different outcomes — which is the point. Claude Code doesn't tell you which one will hold. It tells you the verified facts fast enough that you can make that call yourself.

Reading the Broader Data Claude Code Can Now See

Once connected, the model reaches well past single-ticker lookups — it can read macro proxies, sector rotation, filing flow, and pattern completion, which is where the real edge lives. A few live reads worth knowing you can pull:

Macro backdrop. The current call is Small-Cap Leadership — small caps outperforming large caps, which historically gives small-cap breakouts better follow-through. The proxies: S&P 500 (SPY) $765.96, -1.7% from its 52-week high, 20-day -0.9%. Nasdaq 100 (QQQ) $718.36, -4.0% from its high, 20-day -0.3%. Russell 2000 (IWM) $294.67, -3.4% from its high, 20-day -1.8%. IWM is the small-cap tell; ask Claude Code for it before you size into a runner.

Volume spikes with full session data. The scanner surfaces the highest-RVOL names with all three sessions intact:

Note the mix of green and red closes. DSS ran $0.60 to $1.18 (+106.6% MFE) and closed +39.4% on Sep 3 — and its catalyst string is a lesson in itself: on Sep 4 it announced a proposed public offering, then the same day announced a decision not to proceed, then on Sep 8 launched DSS Robotics. PPBT closed -17.5% despite a +67.6% MFE. A trader who only reads the close misses that both offered a real intraday window.

Sector rotation. Week-over-week average RVOL shifts show where capital is moving: Paper rotated in hard (RVOL 6.03 to 133.85, +2,119%), Communication Services (0.71 to 8.61, +1,116%), Electrical Equipment (1.08 to 3.22, +197%), and Healthcare (6.06 to 15.15, +150%). Ask Claude Code "which sectors are rotating in" and it returns exactly these deltas.

Filing flow. In the past 3 days, the SEC filing corpus logged 171 8-K filings across 165 unique tickers, 4 fresh S-1 registrations (CDT, ACXP, EMAT, OPTT), 2 S-3 shelves (GWH, MLSS), and a single 424B5 pricing supplement (DFDV). Insider Form 4 clusters stacked up too — BTU with 13 filings in 3 days, LMAT with 9, CASH and KTCC with 8 each. Claude Code can list these and flag which clusters are buys versus sells.

Dilution facilities. Across the active universe the server tracks approximate counts (exact totals withheld): ~6,000 active warrant facilities, ~3,200 shelves, ~2,100 ATM programs, ~1,500 convertible notes, ~900 convertible preferred, ~700 S-1 offerings, and ~500 equity lines. When you ask about a specific ticker, Claude Code returns that name's active facilities — the overhang that turns a squeeze into a fade.

Pattern follow-through. The high-volume breakout pattern (stocks trading over 100 million shares intraday) shows 100% follow-through across 118 triggers, with 6 firing this week against a 90-day weekly average of 27.6. Stocks that doubled intraday from session low to high also completed 100% of the time across 201 triggers. Across all pattern types, 62 were detected in the past 7 days at a 100% completion rate — below the 90-day weekly average of 162.7, so this stretch is quieter than normal. Our full method is in Pattern Recognition for Penny Stocks.

Common Pitfalls

The biggest mistake is treating Claude Code like a stock-picking oracle. It is a data-retrieval and reasoning layer, not a signal. When you ask "should I buy GCDT," you'll get a hedge, and rightly so — the model has no business predicting a $0.40 name's next tick. Ask instead for the facts: session OHLC, RVOL, catalyst, active dilution facilities, cash tier. Then you make the call. Prompts that request data get sharp answers; prompts that request predictions get mush.

The second pitfall is ignoring session context. A model — or a trader — that reads only the regular-hours close misses the story. GCDT closed $0.94 but faded to $0.71 after hours. NUR opened $2.02, ran to $7.72, and closed $2.43 the same session (Sep 8) — a +282.2% MFE that gave almost all of it back into the close. If your query doesn't ask for premarket, regular, and after-hours separately, you're reading a third of the tape. Always request all three sessions.

Third: forgetting the dilution check on a low-float squeeze. The names that run 100%+ on huge RVOL are precisely the ones where a company files to sell into strength. LHSW is the cautionary tale — it announced the closing of an $11 million best-efforts follow-on offering on Sep 3, and on Sep 8 the regular session ran $4.50 open to $5.46 high before collapsing to a $1.32 close, -70.7% on the day (TRUE MFE was still +506.7% from the full-day low). The offering was public. Any query that included "recent offerings" would have flagged the supply. Never chase a vertical move without asking Claude Code what's registered.

Fourth: over-trusting stale prices. If you don't confirm the MCP server shows connected, Claude Code may fall back to its training memory and confidently quote a price that's months old. Run claude mcp list and verify. Live data beats a plausible hallucination every time.

How to Apply This

Use Claude Code as the fast front-end and the SNACS scanner as the visual confirm. The workflow that catches setups before they run: ask Claude Code for names trading above 5x RVOL in a rotating-in sector, then open those in the scanner to watch the tape live. Set your scanner RVOL filter to 5x minimum, price $0.50–$20, and add the Dilution Alerts column so overhang shows inline. Click any ticker to open the ticker details page — chart, dilution risk panel (active shelf / ATM / warrant facilities), recent news, and SEC filings, all without leaving the scan.

For the dilution read, you have two paths to the same truth: the scanner's Dilution Alerts column for a glance, and the SEC research dilution snapshot for the full breakdown — active facility counts, shares at risk, lowest exercise price, and the DVS score. When Claude Code flags an active ATM on a runner, confirm it in the dilution snapshot before you size.

To turn a repeatable pattern into an alert, build it in the AI Playbook Builder. Describe the setup — say, "stock trading over 100 million shares intraday with a fresh 6-K catalyst" — and it generates the detection logic; active playbooks monitor every scanner ticker and drop a star indicator on a match. We walk through that in Describe a Trading Setup in Plain English.

Then close the loop in the trading journal. Its AI Insights analyzes your actual fills — best setups, worst time-of-day, and your MFE capture rate. That last metric is the honest scorecard: GCDT offered +224.9% MFE, but what did you capture? If the journal shows you consistently leaving 80% of the MFE on the table, the fix is exit discipline, not more scanning.

Conclusion: What to Watch Next

The setup is one command; the edge is the habit. Traders who wire Claude Code into live data stop guessing about prices and start asking better questions — spike, catalyst, dilution, in one turn. With the macro call sitting at Small-Cap Leadership and IWM within 3.4% of its 52-week high, the small-cap tape stays favorable for follow-through. Watch the sectors rotating in — Paper, Communication Services, Electrical Equipment, Healthcare — and watch the filing flow for fresh S-1s and 424B5 pricings that mark supply hitting a runner. Next session, before you chase anything vertical, run the three-part query: what's the session data, what's the catalyst, and what's registered to sell.

FAQ

What is MCP and why does it matter for connecting Claude Code to market data?

MCP (Model Context Protocol) is an open standard that lets an AI client like Claude Code call external tools and data sources in a consistent format. It matters because it turns Claude Code from a model answering stock questions off frozen training data into a client that queries live scanner, dilution, and SEC-filing data on demand — so its answers reflect today's tape, not last year's.

How do I connect Claude Code to live stock market data?

Run a single command in your terminal: claude mcp add snacs --transport http https://mcp.snacs.trade. This registers the SNACS MCP server under the name snacs. Restart your Claude Code session, confirm with claude mcp list that it shows connected, and the market-data tools are available on your next prompt.

What data can Claude Code access through the SNACS MCP server?

The same universe behind the SNACS scanner: 2,500+ tickers with premarket, regular-hours, and after-hours OHLC, volume, RVOL, float, market cap, and cash runway, plus the SEC filing corpus (1.4M+ filings) and the dilution facility database. That means Claude Code can pull a ticker's session data, its catalyst, and its active shelf/ATM/warrant facilities in one chained query.

Will Claude Code pick stocks or predict prices for me?

No, and you shouldn't ask it to. Claude Code is a data-retrieval and reasoning layer, not a signal generator. Prompts that request facts — session OHLC, RVOL, catalyst, active dilution facilities — get sharp, verified answers. Prompts that request predictions get hedged mush, because the model has no reliable basis to forecast a micro-cap's next move.

Why does session data (premarket vs. regular vs. after-hours) matter in these queries?

Because the regular-hours close alone hides the real move. GCDT closed the regular session at $0.94 on Sep 8 but faded to $0.71 after hours, and NUR ran $2.02 to $7.72 intraday before closing $2.43 the same day. If your query doesn't request all three sessions, you're reading a fraction of the tape and misjudging both the opportunity and the overnight risk.

How does the MCP connection help me avoid dilution traps?

The server exposes each ticker's active dilution facilities — shelves, ATMs, warrants, convertible notes — so Claude Code can flag supply before you chase a vertical move. LHSW announced the closing of an $11 million follow-on offering on Sep 3, then ran to a $5.46 high on Sep 8 before collapsing to a $1.32 close (-70.7%). A query that included recent offerings would have surfaced that supply first.

Does this work with editors other than Claude Code?

Yes. The same SNACS MCP server connects to VS Code Copilot, Cursor, and the Claude web and desktop apps — the protocol is identical, only the add command syntax differs per client. Each has its own setup guide, but the underlying data and tools are the same across all of them.

How do I know the market data is live and not the model's training memory?

Run claude mcp list and confirm the snacs server shows connected. If it isn't connected, Claude Code may fall back on stale training data and quote a months-old price with full confidence. When the server is connected, the model calls the live endpoint and returns current session data — always verify the connection before trusting a number.