FMP Stock Market API

TL;DR

Before choosing FMP, compare it with seven alternatives in the free stock API comparison.

Who this FMP API review is for

Start with the current Stable API contract

FMP's current Quickstart documents https://financialmodelingprep.com/stable/ as the base URL. It lists API-key authentication, more than 100 endpoints, and common 403, 429, and 500 responses. Put the base URL, authentication mode, timeout, and plan limits in configuration instead of scattering them across notebooks and prompts.

FMP stock market API coverage map

The useful unit is not “an endpoint.” It is a dataset with an identity, field contract, update cycle, adjustment policy, geography, history window, and permitted use.

Free tier, paid limits, and licensing reality

FMP's public pricing page currently separates personal and commercial use and lists different call rates, history, coverage, bandwidth, and datasets by plan. Prices and packages can change, so this page does not hard-code a procurement quote. Capture the values on the day you evaluate and attach the applicable agreement.

Code: call the FMP Stable API with cURL

Use an environment variable and avoid exposing the key in shell history or screenshots. FMP documents both header and query-parameter authorization; a header keeps secrets out of the URL.

curl --fail-with-body \
  --connect-timeout 5 \
  --max-time 30 \
  -H "apikey: ${FMP_API_KEY}" \
  "https://financialmodelingprep.com/stable/historical-price-eod/full?symbol=AAPL"

Python: retry, normalize, and preserve evidence

A production adapter should return a provider-neutral record and an evidence envelope. The example retries only transient failures, respects a deadline, records retrieval time, and keeps the raw response hash without placing the API key in logs.

import hashlib, json, os, random, time
from datetime import datetime, timezone
import requests

BASE_URL = "https://financialmodelingprep.com/stable"
API_KEY = os.environ["FMP_API_KEY"]

def get_eod(symbol: str, attempts: int = 4) -> dict:
    url = f"{BASE_URL}/historical-price-eod/full"
    headers = {"apikey": API_KEY}
    params = {"symbol": symbol}

    for attempt in range(attempts):
        response = requests.get(
            url, headers=headers, params=params, timeout=(5, 25)
        )
        if response.status_code == 403:
            raise RuntimeError("FMP authentication or entitlement rejected")
        if response.status_code not in {429, 500, 502, 503, 504}:
            response.raise_for_status()
            break
        if attempt == attempts - 1:
            response.raise_for_status()
        time.sleep(min(8, 0.5 * (2 ** attempt)) + random.random() / 4)

    raw = response.content
    rows = response.json()
    normalized = [{
        "symbol": symbol,
        "date": row["date"],
        "open": row["open"],
        "high": row["high"],
        "low": row["low"],
        "close": row["close"],
        "volume": row["volume"],
        "provider": "fmp",
        "dataset": "historical_price_eod_full",
    } for row in rows]

    return {
        "records": normalized,
        "evidence": {
            "retrieved_at": datetime.now(timezone.utc).isoformat(),
            "request": {"endpoint": "/historical-price-eod/full", "symbol": symbol},
            "status": response.status_code,
            "sha256": hashlib.sha256(raw).hexdigest(),
        },
    }

Adjusted prices, timestamps, and point-in-time truth

FMP exposes full, non-split-adjusted, and dividend-adjusted historical price endpoints. These series answer different questions. Record the endpoint and adjustment mode with every dataset; never silently join adjusted closes to unadjusted highs and lows.

• Store event time, provider source time, retrieval time, and processing time separately.
• Test pre-market, regular, after-hours, weekend, holiday, and daylight-saving boundaries.
• Detect duplicates, reversed ranges, negative volume, missing sessions, and impossible OHLC relationships.
• Pin a dataset version or raw-response hash so an agent answer can be reproduced.

A data-quality acceptance contract

dataset: fmp_historical_eod
symbols: 100
window: 2025-01-01..2026-06-30
contract:
  required_fields: [symbol, date, open, high, low, close, volume]
  adjustment_mode: documented
  currency_and_exchange: resolved
  duplicate_keys: 0
  impossible_ohlc_rows: 0
  missing_expected_sessions_pct: "<= 0.25"
  stale_quote_rate_pct: "<= 0.50"
  p95_latency_ms: "<= 2500"
  unsafe_secret_logs: 0
evidence:
  raw_hash: required
  retrieved_at: required
  provider_endpoint: required
  validation_result: required

Set thresholds from your workload and market calendar; the numbers above are examples, not universal guarantees. Release only after replaying a frozen symbol/date set and reconciling a sample against an authoritative source.

Production architecture for FMP data

Troubleshooting FMP API failures

Compare FMP with one shared workload

Do not compare providers with different symbols, dates, plans, caches, and acceptance rules. Replay the same frozen workload and score accepted datasets.

Migration checklist for endpoint or provider changes

• Inventory every endpoint, caller, key location, quota, cache key, schema, adjustment assumption, and downstream consumer.
• Freeze representative responses and expected normalized records before changing the adapter.
• Run old and new paths in read-only shadow mode; compare values, dates, missing rates, latency, and cost.
• Canary one low-risk dataset, preserve rollback, and block writes or agent conclusions when evidence is incomplete.
• Update runbooks, dashboards, licensing records, secret rotation, and data-retention controls before cutover.

QVeris pattern: discover, inspect, call, then validate

QVeris fits when an existing agent needs a capability layer rather than a hard-coded provider endpoint. The agent can discover a financial capability by intent, inspect parameters and operational signals, call the selected capability, then feed only validated evidence into reasoning.

FMP stock market API FAQ

Official sources and update policy

Endpoint, pricing, freshness, and licensing claims were checked against official sources on July 31, 2026. Recheck the current plan and agreement before procurement or release.