Skip to main content

Financial data API for AI agents

Most market-data APIs were designed for human dashboards: one ticker per call, REST endpoints scoped to a single domain (quotes, fundamentals, filings, news), and monthly subscriptions paid by a person with a credit card. Agents work differently. A single research turn often needs five tickers across four domains, and the agent itself — not a human — is paying. Jintel is shaped around those constraints.

What changes when the caller is a model​

There are three shifts that matter:

  1. Typed schemas beat freeform JSON. Agents call tools through structured schemas (OpenAI function calling, Anthropic tool use, MCP). A typed GraphQL schema means the model knows exactly which fields exist, which are nullable, and what shape the response will take — without scraping a Markdown reference page. The full Jintel schema is discoverable via standard GraphQL introspection and the agent-friendly tool catalog advertised on /openapi.json.

  2. Fan-out beats N+1. A "should I buy NVDA" prompt expands into roughly: latest quote, last 30 days of price history, fundamentals, analyst consensus, recent news, technical indicators, and any insider trades. That's six REST calls per ticker on a typical vendor — six round trips, six rate-limit slots, six places where one call can fail and break the chain. With Jintel, every one of those is a sub-graph on Entity and resolves in a single round trip via Mercurius loaders. See the sub-graphs guide for the full list.

  3. Per-query pricing beats subscription tiers. Agents that call APIs from someone else's wallet (or their own budget) need to know the cost of a request before they make it. Jintel exposes a free estimateCredits(query, variables) lookahead and prices each POST /api/graphql call deterministically based on the GraphQL AST. On the x402 rail, the agent's wallet signs a USDC payment per request — no signup, no API key, no monthly contract.

Schema-shape for batched research​

The two queries you'll use most are quotes(tickers: […]) and entitiesByTickers(tickers: […]). Both accept a list and fan out internally. quotes is the cheapest path when all you need is price + change. entitiesByTickers returns full Entity objects whose nested sub-graphs (market, news, analyst, technicals, regulatory, insiderTrades, earnings, ownership, ...) lazy-resolve only what you select. Selecting nothing costs nothing; selecting six sub-graphs across five tickers is one HTTP call.

Crypto tickers route to CoinGecko / Binance automatically — quotes(tickers: ["AAPL", "BTC"]) is valid. Equity tickers route to Yahoo Finance. The mixing is transparent at the schema level.

A research-agent fan-out​

Here's the shape of a single GraphQL request that gives an agent enough context to write a one-paragraph note on each ticker:

query AgentResearch($tickers: [String!]!) {
  entitiesByTickers(tickers: $tickers) {
    name
    tickers
    market {
      quote { price changePercent volume marketCap }
      fundamentals {
        peRatio forwardPE pegRatio priceToSales
        revenueGrowth earningsGrowth grossMargin operatingMargin
        sector industry
      }
    }
    analyst {
      recommendation targetMean numberOfAnalysts
    }
    technicals { rsi ema50 sma200 macd { histogram } }
    news(filter: { limit: 5, sort: DESC }) {
      title date source sentimentScore link
    }
    insiderTrades(filter: { limit: 5 }) {
      reporterName officerTitle transactionDate
      acquiredDisposed shares pricePerShare transactionValue
    }
  }
}

Run it once with { "tickers": ["NVDA", "AMD", "INTC"] } and you get five domains of data on three tickers in a single response, billed against a single per-query x402 quote (or a single deduction from your monthly plan credits).

curl https://api.jintel.ai/api/graphql \
  -H "Authorization: Bearer $JINTEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d @- <<'JSON'
{
  "query": "query($t:[String!]!){entitiesByTickers(tickers:$t){name tickers market{quote{price changePercent} fundamentals{peRatio sector}} analyst{recommendation targetMean} news(filter:{limit:3}){title date sentimentScore}}}",
  "variables": { "t": ["NVDA", "AMD", "INTC"] }
}
JSON

Where to go next​

  • Sub-graphs — full list of nested fields available on Entity and MarketQuote.
  • Batching — how Mercurius loaders deduplicate fan-out and why a single request beats N requests.
  • Agents & x402 — pay-per-query with an agent wallet, no signup.
  • MCP server — drop Jintel into Claude Desktop or any MCP-aware client.