How to Add Live Market Data to Your Claude Code Agent
Give Claude Code a live market data feed in about two minutes: one free API key and one command, and your agent can screen stocks, read sentiment, check who covers a name, and see who moved after earnings, instead of guessing from training data.
To add live market data to Claude Code, get a free SentiSense API key and run one command: either add the hosted MCP connector, or install the SentiSense skills. From there the agent can screen stocks, read sentiment, pull analyst coverage, check insider and institutional activity, and look up the earnings calendar, without you writing an integration.
We run Claude Code this way every day, on the same data we sell, so everything below is a path we use ourselves, not a demo.
What does Claude Code know about markets on its own?
Nothing current. Ask an unconnected agent about a stock and it answers from training data that is months old, or it reaches for a web search and pastes whatever headline came back. For code that is fine. For a question like "which firms moved after the print" it is worse than nothing, because the answer arrives in a confident voice.
A data feed changes the shape of the answer. The agent calls a tool, the tool returns numbers with dates on them, and the agent reads them back. When it says "4 of 37 covering firms moved in the five sessions after the report", that sentence came out of a response, not a guess.
How do you set it up?
Two minutes, and you only need one of the two paths.
Path 1: the hosted connector (MCP). Best when you want the agent to pick tools itself. From your project directory:
claude mcp add sentisense -- npx -y mcp-remote@0.1.38 https://app.sentisense.ai/mcp
Start claude, approve the server when it asks, and sign in with your SentiSense account in the browser tab that opens. That is the whole install. The connector is read-only and exposes ten data tools, from a market-wide mood composite to analyst ratings, options positioning, 13F flows, and a screener. The connector guide lists every tool.
Path 2: the skills. Best when you want the agent to follow a documented playbook per task, and it works in Codex, Grok Bot and Gemini too:
npx skills add SentiSenseApp/skills
Then put your key in the environment as SENTISENSE_API_KEY. Get one free at sentisense.ai/get-api-key. There are nineteen skills; the onboarding one routes the agent to the right skill for whatever you typed, so you do not have to remember names.
You can run both. The connector answers quick questions in a chat; the skills carry the longer workflows and fixed output templates.
What can it actually do?
Everything below is something our own Claude Code sessions did this week, given only the question:
- "Who covers NVDA and where does each firm stand?" One call, 41 firms, bucketed into Buy, Hold and Sell, with the named analyst on each row.
- "Which firms moved after CRM's print?" "4 of 37 covering firms moved in the five sessions that followed... 33 firms have not published since."
- "Do analysts and the crowd agree on HOOD?" The Street's rating split next to the news and social read on the same name, in one sentence.
- "Screen for large caps with a rising SentiSense Score."
- "Is the insider selling in this name conviction selling or planned 10b5-1 sales?"
- "What is on the earnings calendar next week for my watchlist?"
The feed behind these covers sentiment, analyst, insider, institutional, options and earnings data for roughly a thousand of the most-watched US stocks. It is read-only by design: the agent can look things up, and nothing more.
The delegation pattern
Claude Code's subagents make the setup compose. Give the feed to the session once, then let the planning model hand market questions to a cheaper worker: a scout that pulls coverage and consensus for a list of tickers, a writer that turns the numbers into a note, a checker that re-derives one figure from a second endpoint before anything ships. We run that loop with the planner holding the thread and the workers holding the data, and the second-opinion worker, the one with no context of the plan, is the one that catches the mistakes.
The same skills work in Codex, Grok Bot, Gemini and OpenClaw, so the analyst moves with you if your runtime does. Data is provided for research and education, not investment advice.