Free crypto price APIs can serve production reads if you treat them as a cached, rate-limited source instead of a live feed. Budget monthly credits first, then poll no faster than the provider's documented refresh interval. Reject snapshots older than your staleness limit, and keep one backup provider ready for failover.
- Free tiers usually cap calls per minute and per month, so one hot loop can exhaust the whole budget.
- Keyless endpoints are often shared by IP address, so other users can push you into 429 responses.
- Store the provider timestamp with each snapshot, and alert when the newest value passes your staleness limit.
Where a Free Price Feed Breaks in Production
Most teams pick a free crypto prices API without a budget model, then meet the failure boundary weeks later. The price value itself is rarely the problem. The limits are.
CoinGecko's documentation lists its authenticated Demo plan at 100 calls per minute, and its keyless, no-signup endpoints sit in a shared IP pool of roughly 10 to 30 calls per minute. CoinMarketCap documents a free Basic tier with 15,000 monthly credits at 50 requests per minute. These limits were checked on September 29, 2026, and they change often, so confirm them again on the provider's page before you deploy.
Data Flow: Which Component Owns the Snapshot
Give one collector process the only provider credentials, and let every downstream service read the collector's store instead of calling the provider directly. That single rule turns a shared, metered resource into a controlled one. The collector writes a row per symbol with the provider's own update time, your fetch time, the price, and the provider name.
Your price service then answers reads from that row and can state how old the value is. Two timestamps matter because a fresh fetch of an old provider value is still stale data.
Implementation Choices, Failure Modes, and Recovery Controls
Poll no faster than the provider's documented refresh interval, because extra calls spend credits without adding new information. When a 429 response arrives, back off with jitter instead of retrying at once, and hold a circuit breaker so one slow provider cannot pile up requests. Keep a second provider behind the same interface, and tag stored rows with the source so downstream users can see when a switch happened.
- Rate limit exhaustion: cap in-flight requests, add exponential backoff with jitter, and honor retry hints the provider returns.
- Silent staleness: reject or flag any snapshot older than the freshness limit your strategy needs.
- Symbol mismatch: map provider IDs rather than tickers, since two assets can share one ticker.
- Mid-month credit exhaustion: alert at a set share of the monthly budget, not at zero.
Verification, Observability, and Go-Live Criteria
Track four numbers before launch. Calls consumed against the provider's published budget, the share of responses that are 429s, the cache hit ratio, and the age of the newest snapshot at each consumer. Run the collector for a week at peak load and confirm leftover budget; if peak usage sits near the cap, the design will fail on the first traffic spike. Test failover by pointing the collector at the backup provider in staging and comparing the same symbols.
Free REST snapshots fit dashboards, risk pre-checks, and reference pricing. They do not fit order matching or fill pricing, where you should read the exchange's own trade stream.
Sources
- Errors & Rate Limits CoinGecko · 2026-09-29
- Keyless Public API CoinGecko · 2026-09-29
- CoinMarketCap API for AI Agents CoinMarketCap · 2026-09-29
- The Free Crypto API Pricing Stress Test: What Happens When a Product Starts Shipping CoinMarketCap · 2026-09-24 · 2026-09-29