For agents

Connect your agent

StockIntel runs next to the Binance server inside your own agent session. Add us once, then ask for market reads, theses, evidence, or a fresh live investigation. Read-only, no key needed.

Quickstart

Your agent can read the market in one request

Plain HTTP, JSON in, JSON out. Install the skill and any Agent OS client, Claude Code, Codex, ChatGPT, can call it by name.

1, a live market read

curl -s -X POST https://stockintelislive.vercel.app/api/market/read \
  -H "Content-Type: application/json" -d '{"tickers":["NVDA"]}'

2, the thesis behind it

curl -s -X POST https://stockintelislive.vercel.app/api/market/assess \
  -H "Content-Type: application/json" -d '{"ticker":"NVDA"}'

3, or run it as an MCP server, beside Binance's own

claude mcp add binance-mcp-server --transport http https://agent.binance.com/mcp/agentic
claude mcp add stockintel --transport http https://stockintelislive.vercel.app/mcp

Claude Code

claude mcp add stockintel --transport http https://stockintelislive.vercel.app/mcp

Codex

codex mcp add stockintel --url https://stockintelislive.vercel.app/mcp

VS Code

Open this repo. Both servers start from .vscode/mcp.json, then MCP List Servers, authenticate the Binance one.

Cursor

Settings, MCP, add https://stockintelislive.vercel.app/mcp as a Streamable HTTP server.

ChatGPT on web

Settings, Security, turn on Developer Mode and log in. Open Plugins, press +, name it StockIntel, paste https://stockintelislive.vercel.app/mcp as the plugin URL, Create.

What a thesis looks like

The grid observes and hypothesizes. The orchestrator connects the dots into one assessment per exposure. Example shape below, illustrative, not a live call.

01 · Assessment · Not yet priced

Hyperscale buildout → GPU demand → NVIDIA → TSMC capacity

A $10B data-center expansion fans out through GPUs, memory, servers, and power. The chain reaches NVIDIA through confirmed procurement, and the observed price move does not yet reflect it. Breaks if the buildout slips a quarter or the order goes elsewhere.

Market call · timeframe 1 to 4 weeks · confidence 82%

What each call does

  • stockintel_investigate

    Starts the full ten-specialist grid on your question, returns an inquiry id.

  • stockintel_inquiry

    Polls that investigation to the finished thesis.

  • stockintel_assess

    The latest stored thesis for a ticker, priced or not, with market calls.

  • stockintel_clusters

    Grouped evidence per exposure, no verdicts.

  • stockintel_thesis_changes

    What moved between the last two assessments, plus history.

  • stockintel_conflicting

    What argues against the latest thesis, from stored data only.

  • stockintel_evidence

    Drill one thread down to sources, cluster, and top claims.

  • stockintel_read

    Live prices, 24h move, and the positioning gauge for up to 20 tickers.

  • stockintel_status

    Whether this deployment holds its own Agent OS session.

Prefer plain web requests? Every tool above has an HTTP twin under /api/market, and your own Binance numbers ride along as caller-supplied context. The full contract lives in skills/stockintel/SKILL.md in the repo.

Check the connection works

One prompt proves it

In your agent chat, enter: use the StockIntel MCP server to read NVDA. Confirm the response shows the StockIntel tool being used and returns a live read. For the combined proof, ask it to fetch the BTC ticker from your Binance tools first and pass that output into the StockIntel read.