How to Add Live Market Data to Your Codex Agent
Give Codex a live market data feed in about two minutes: one free API key and one command, and GPT 6 Astra can screen stocks, read sentiment, check who covers a name and see who bought after earnings, instead of guessing from training data.
To add live market data to Codex, get a free SentiSense API key and register the SentiSense MCP connector with one command. From there, Codex can screen stocks, read sentiment, pull analyst coverage, check insider and institutional activity and look up the earnings calendar, without you writing an integration or pasting CSVs into the prompt.
We run Codex this way every day, on the same data we sell, so everything below is a path we use ourselves, not a demo.
What does Codex know about markets on its own?
Codex knows what its training data knew, up to a cutoff that is always months in the past. Ask it what an analyst's price target is on a stock, whether insiders bought after the last earnings print or how sentiment shifted this week, and you get one of two answers: a confident number that is stale, or a polite refusal. Astra reasons well about what it is given. It has no way to fetch what it is not given.
That is the gap a market data connector closes. The model keeps doing the reasoning; the connector supplies the facts.
How do you connect Codex to live market data?
There are two paths. Use the MCP connector when you want the agent to call the data directly, and use the skills when you want repeatable, documented workflows the agent can follow.
Path 1: the MCP connector. Register the SentiSense server in the Codex CLI and sign in once in your browser:
codex mcp add sentisense --url https://app.sentisense.ai/mcp
codex mcp login sentisense
In the Codex desktop app, add the same server under Plugins, then Manage, then MCP, then Add server, by URL: https://app.sentisense.ai/mcp. The app and the CLI share one configuration, so a server added in either is available in both.
Codex now has the SentiSense tools available in every session: stock snapshots, screening, sentiment, analyst ratings, earnings, insider and 13F activity, options and the market mood. Ask a question in plain language and the agent picks the tool.
Path 2: the skills. Ask your agent to install them: tell Codex to run npx skills add SentiSenseApp/skills, or run it yourself:
npx skills add SentiSenseApp/skills
The installer places the SentiSense skills in Codex's own skills folder (.agents/skills/ for the current project, or add --global to make them available everywhere), and Codex loads them at the start of every session, with nothing to add to AGENTS.md. Each skill is a Markdown playbook for one workflow, such as a screener run, an earnings read or a 13F comparison, with the exact endpoints, the output template and the caveats. The full list, with what each one does, is on the agent skills page; OpenClaw users get the same skills from ClawHub. The skills call the REST API with your free key, so set SENTISENSE_API_KEY in your shell first. Get the key at sentisense.ai/get-api-key. No card, and a free key is enough to try everything in this post.
What can Codex actually do with it?
The first prompts people try, and what comes back:
- "Screen for stocks with rising sentiment and a SentiSense Score above 70." The agent runs the screener and returns a ranked table with the score, the 7-day sentiment change and the mention trend for each name.
- "Who covers NVDA and what is the consensus target?" Analyst coverage with the named firms, their latest targets and the direction of recent changes, not a single blended number.
- "Did insiders or institutions buy after the last earnings print?" Form 4 trades and the newest 13F changes for the ticker, dated, so the agent can say who moved and when.
- "What did the market read into last week's earnings?" The ranked earnings reactions, the surprise versus estimate and the sentiment shift in the days after the print.
Each answer carries its source and timestamp, so when Astra reasons about the numbers you can see exactly which numbers it reasoned about.
The second-opinion pattern
The most useful thing we do with Codex is not research from scratch. It is the second opinion. We build a thesis or a screen in one agent, then hand the same question to Codex with the connector attached and ask it to verify every figure against the live data and flag what does not hold. Two models reading the same live feed disagree in useful places; two models reciting training data agree on the same stale story. If you run more than one coding agent, this is the cheapest way to get a real second opinion on a market question. We wrote up how the two agents split the work in our Claude Code vs Codex comparison.
Questions people ask
Does this work with GPT 6 Astra in Codex? Yes. The connector and the skills are model-agnostic. Astra is the model most people run Codex with today, and it is a strong reviewer of numbers once it has live numbers to review.
Do I need a card to try it? No. A free SentiSense key covers everything in this post, with a monthly request allowance and a per-minute rate limit. Paid plans lift the limits.
Does Codex read CLAUDE.md? No. Codex reads AGENTS.md for project instructions and loads skills from its own skills folder. The installer puts the SentiSense skills there, so there is nothing to copy by hand.
Is the data live or delayed? Sentiment is refreshed several times a day as new coverage is processed. Analyst coverage, insider filings, 13F holdings and fundamentals update as the filings land. Quotes can be delayed by up to 15 minutes. Every response carries its own timestamp so the agent can tell you how fresh a number is.