Connect
How to connect ChatGPT to live markets data
ChatGPT now supports MCP connectors through Developer Mode, which means you can wire it directly to a structured markets database. Here is the exact setup — including the URL-embedded-key trick that gets around the missing header field in the connector dialog.
The short version
- ChatGPT’s Developer Mode lets you attach MCP connectors to any chat.
- The connector dialog offers only OAuth or “No authentication” — no header field. The fix is to embed the key in the URL path instead of a header.
- Get a free
sk-livekey at /setup, build the URLhttps://mcp.clawterminal.ai/k/sk-live-.../mcp, and add it with auth set to None. - Once connected, ask in plain English: recent 8-Ks, insider buying clusters, trailing fundamentals, FRED macro series, top movers.
Why this is worth doing
ChatGPT without live data tools is a strong reasoner working from a frozen snapshot. Its training cutoff means it can describe how to analyze a 10-K in detail but cannot tell you what was actually in a company’s most recent one. Ask it for a specific insider transaction price and it may produce a plausible-looking number that has no connection to what was filed. The underlying capability is real; the data is not.
MCP connectors change that. Instead of asking the model to remember market data, you give it tools that retrieve it. The model decides when to call a tool and how to interpret the result; the server handles the data fetch. The answer is now anchored to a real query against a real database, not a confident reconstruction from training weights.
For markets research that means you can ask ChatGPT to compare insider buying across two tickers, or to find 8-K filings that mention a CEO departure this month, or to check how a stock reacted to its last earnings report — and get results that reflect actual filed documents rather than approximations.
The authentication problem in ChatGPT’s connector dialog
Before the setup steps, it is worth understanding the one wrinkle that catches almost everyone. ChatGPT’s connector dialog offers two authentication options: OAuth and “No authentication”. There is no field for a custom Authorization header.
That is a problem because the standard way to authenticate against an MCP server is a Bearer token in the Authorization header — the same pattern used by almost every API. Without that field, there is no obvious way to pass your key.
The solution is the URL-embedded-key route. Instead of sending the key as a header, you embed it in the path itself:
https://mcp.clawterminal.ai/k/sk-live-.../mcpThe server’s reverse proxy (Caddy) intercepts requests to that path, strips the /k/<key> prefix, and reinjects the key as a standard Authorization: Bearer <key> header before forwarding the request inward. The MCP server itself sees a normal authenticated request and validates it identically to any other key. The authentication is real; only the delivery channel is different.
The URL is the credential. Once your key is embedded in the URL, that URL has the same threat model as a Bearer token: anyone who sees it can use it. Never paste the full URL in a shared chat, a public screenshot, or a version-control commit. If the URL is ever exposed, go to /setup, issue a new key, and deactivate the old one immediately.
Step 1 — get a free key
ClawTerminal is a markets-research MCP server in free closed beta. Sign in at /setup with your email address — a magic link is sent, no password needed. After you verify, your personal sk-live-… key is displayed. Copy it and keep it somewhere safe; it is shown once.
Once you have the key, build your connector URL by inserting it into the path:
https://mcp.clawterminal.ai/k/sk-live-YOUR-KEY-HERE/mcpThis is the URL you will use in step 3. Do not share it.
Step 2 — enable Developer Mode in ChatGPT
MCP connectors are gated behind Developer Mode, which is off by default. To turn it on:
- Open ChatGPT and go to Settings.
- Select the Advanced tab.
- Scroll to Developer Mode and enable it.
You only need to do this once. Developer Mode persists across sessions.
Step 3 — add the connector
With Developer Mode enabled:
- Go to Settings → Connectors → Create.
- Fill in the fields:
Name —ClawTerminal
URL —https://mcp.clawterminal.ai/k/sk-live-…/mcp(your full URL from step 1)
Authentication —No authentication - Save the connector.
Setting authentication to “No authentication” is correct here — the key is already in the URL, so you are not sending a second credential on top of it. The server will authenticate via the path prefix regardless of what the dialog sends for auth.
Step 4 — use it in a chat
Connectors are attached per-chat rather than globally. To use ClawTerminal in a conversation:
- Open a new chat (or an existing one).
- Click the + button in the message input area.
- Select Developer Mode → choose ClawTerminal.
- Ask your question in plain English. ChatGPT selects and calls the appropriate tools automatically.
You do not need to name tools or write queries. If you ask “what insider buying happened at biotech companies last week”, ChatGPT identifies the right tool, calls it with the right parameters, and synthesizes the result into a readable answer.
Using Claude Desktop? The same URL-embedded-key route works in Claude Desktop’s custom-connector dialog, which has the same header-field limitation. However, if you are setting up Claude Desktop or Claude Code directly (not through the custom-connector dialog), use a normal Authorization: Bearer sk-live-… header — that path has no limitations. See our Claude setup guide for the full config.
Example prompts — what you can ask
Here are representative questions along with what the server actually returns. You never need to write these as structured queries; plain English works.
| You ask ChatGPT | What it pulls |
|---|---|
| “Show the most recent 8-Ks for Tesla.” | Dated list of 8-K current reports with item codes, filing dates, and links to the underlying documents |
| “Where have insiders been buying shares in clusters recently?” | Form 4 open-market buy clusters across tickers: insider roles, transaction dates, total values, and price ranges |
| “What are Microsoft’s trailing twelve-month revenue and free cash flow?” | XBRL companyfacts series, TTM-aggregated from SEC-filed GAAP data |
| “Pull the 10-year Treasury yield and CPI series from FRED.” | Macro observations with dates and values from the FRED public-domain dataset |
| “Which US equities moved the most today?” | Top movers ranked by percent change with price, volume, and sector |
The server handles entity resolution automatically. If you type a ticker, a company name, or a CIK, the resolve_entity tool maps it to a canonical identifier before calling downstream tools. You do not need to know whether to pass a ticker or a CIK; ChatGPT chains the resolution step itself.
What is available once connected
ClawTerminal is a full markets terminal, not just an SEC connector. The surface you get with a single key includes:
- SEC filings — 8-K, 10-K, 10-Q, Form 4 insider transactions, Schedule 13D/13G beneficial ownership, 13F-HR institutional holdings, N-PORT fund holdings, and DEF 14A proxy statements. All semantically searchable, with price-reaction studies available for events.
- Fundamentals — XBRL companyfacts aggregated to TTM and point-in-time, with multi-company comparison and over 100 concepts per issuer.
- Prices — split-adjusted daily OHLCV and dividends for US equities, with top-movers and correlation tools.
- Macro — 104 FRED public-domain series including yields, CPI, employment, GDP components, and FX rates. Plus USDA crop data, EIA energy series, and World Bank commodity prices.
- Crypto — Hyperliquid perps across core coins and builder dexes, including pre-IPO private equity (OpenAI, SpaceX, Anthropic via Ventuals) and traditional assets (crude, gold, equity names via trade.xyz).
- Prediction markets — cross-venue event odds from Polymarket and Kalshi.
- Equity screener — multi-rule filter and sort over ~5,000 US issuers with TTM fundamentals, price ratios, and trailing risk metrics.
That is over 160 tools behind one endpoint. Our full data-surface overview goes deeper on what each domain covers and where the data comes from.
A note on data quality and interpretation
Connecting ChatGPT to a real database is a significant upgrade over relying on training-data recollection. But the tool returning the correct number does not make the analysis correct. A few things to keep in mind:
- 13F holdings are filed 45 days after the quarter ends — the positions you see are stale by design.
- Congressional trade disclosure values are bucket midpoints, not exact dollar amounts.
- Insider buying clusters include noise from IPO lockup expirations and PIPE transactions that look like conviction buys but are not.
- Fundamentals from XBRL are as filed; restatements appear as new rows rather than corrections to old ones.
The data is real. The interpretation still requires judgment. ChatGPT will surface the right filings; you decide what they mean.
That’s the setup
Three steps: get a free key, enable Developer Mode, add the connector with the URL-embedded key. From that point on, any chat you attach ClawTerminal to has access to a live, structured markets database. The model stops guessing at specific filing details and starts retrieving them. That is the shift worth making.