HTX trading API vs KuCoin

HTX trading API 25

The editorial half. A closed service whose user agreement exists but couldn't be read (script-drawn pages), so the licence clarity is unknown rather than unclear (10 of 30). The privacy policy was not read (0 of 30, unchecked). The testnet shutdown and the 2022 changelog entries are dated, with no deprecation policy (5 of 20). No subprocessor or data-location statement read (0 of 20). 10 + 0 + 5 + 0 = 15.

KuCoin 54

The editorial half. A closed hosted service with terms that are clear on use, liability and law but name no company (12 of 30). The privacy policy gives retention periods by purpose and provider (2 years after closure, 14 months for Firebase, 3 years for Jumio, 5 years for Sumsub) and names 41 services, with no data centre locations (22 of 30). Endpoints are removed and fields deprecated in dated changelog entries with no notice period or policy (8 of 20). Third parties named by service, locations not stated beyond transfers around the globe (10 of 20). 52.

HTX trading API 40

No OpenAPI or other machine-readable contract found in the docs or the GitHub accounts (0). No llms.txt on htx.com, and the readable reference is a static page (0). Each endpoint states its purpose and parameters, without when-to-use guidance (12 of 20). Parameter tables carry types, required flags and enums for order types and states (10 of 15). Request and response examples throughout and error-code tables with messages (11 of 15). Versions v1 and v2 are mixed in one reference whose last changelog entry is 24 June 2022, while HTX's newer docs on htx.com couldn't be read (7 of 15). 0 + 0 + 12 + 10 + 11 + 7 = 40.

KuCoin 60

No OpenAPI or other machine-readable contract was found; the docs link a Postman workspace and the universal SDK is generated from an internal spec (8 of 25). No llms.txt; the skills hub ships Markdown SKILL.md files with endpoint tables for seven product areas, read-only (6 of 10). One docs page per endpoint with a description, permission and weight, but no when-to-use guidance (12 of 20). Typed parameter tables with enums for type and side and required flags (11 of 15). Request and response examples per endpoint and a unified six-digit error code scheme introduced with V2 (10 of 15). Dated changelog with V1 to V2 versioning (13 of 15). 60.

HTX trading API 15

No machine payment protocol (0). The fee schedule at htx.com/fee is drawn by script and answered no text, so no price is recorded first-hand (0 of 20, unchecked). Public market data needs no key, card or account; trading needs a verified and funded account. Judgement call (15 of 20). A person must sign up in a browser and pass identity checks (0). 0 + 0 + 15 + 0 = 15.

KuCoin 35

No machine payment protocol (0). The spot base fee is stated in support articles without login, the futures base rate and the full tier table sit in images or behind script (15 of 20). Market data needs no key, no card is asked for at sign-up (20). A person registers, verifies identity and mints the key in a browser (0). 35.

HTX trading API 57

Read as a closed service. The readable changelog ends on 24 June 2022, but the official contract SDKs were pushed on 21 October 2026 and the spot SDKs in April and July 2026, so the API surface is still being maintained; judgement call (20 of 30). SDK pushes on 30 July, 23 August and 21 October 2026 are commits, not releases (10 of 20). A Telegram group for API support and a support centre, neither tested; GitHub issues not read (8 of 15). Official SDKs in five languages (15). No CI seen, dependencies not checked (4 of 10). 20 + 10 + 8 + 15 + 4 = 57.

KuCoin 73

Read as a closed hosted service. The changelog's newest past entry is 2 November 2026 and the universal SDK was pushed 15 September (30). Nine dated changelog entries since 5 August 2026 (20). A Telegram API group (t.me/KuCoin_API) and api@kucoin.com for limit requests, no public issue tracker for the API; the SDK repository's issues were not checked (10 of 25). Official SDKs current in five languages, no MCP registry presence (13 of 15). No package health signal checked (0 of 10). 73.

HTX trading API 38

Read with the hosted lines. No status page found; status.htx.com does not resolve and nothing on htx.com was readable (0). No readable incident history (5). Rate limits documented per endpoint with numbers, 100 order calls per 2 seconds per UID and so on, with remaining-quota headers (15). No 429 or Retry-After behaviour is documented and the only limit error named is 1006 for conditional orders; a client-order-id unique within 8 hours and an idempotent withdrawal id give safe retries on writes (8 of 15). No SLA found (0). The spot and swap APIs are generally available (10). 0 + 5 + 15 + 8 + 0 + 10 = 38.

KuCoin 45

Graded with the hosted lines on the API at api.kucoin.com. No status page found (status.kucoin.com does not resolve, the docs link none), so 0 for the page and 5 for no readable history. Rate limits are published per VIP level with per-endpoint weights and IP limits for public endpoints (15). Every response carries gw-ratelimit-limit, gw-ratelimit-remaining and gw-ratelimit-reset, 429000 is documented with a wait rule, and orders take a clientOid the docs recommend as a UUID (15). No SLA (0). The UTA V2 API is the production surface (10). 45.

HTX trading API 69

Graded on the REST API. Responses can be sized with depth types, size and limit parameters on books, candles and history (15 of 25). Paging by from, size and direct on orders and history (14 of 20). Errors return err-code strings with messages, listed in tables (13 of 20). A client-order-id unique within 8 hours and idempotent withdrawal ids make retries safe, with no idempotency header on other writes (14 of 20). Few required parameters and official SDKs in five languages (13 of 15). 15 + 14 + 13 + 14 + 13 = 69.

KuCoin 62

Read for the API alone, since there is no MCP server. Responses take limit and cursor parameters on list endpoints and the order book depth is a parameter (14 of 25). Pagination and symbol filters on orders, fills and ledgers (15 of 20). Codes and messages per response, with the V2 scheme claimed to be clearer; the rate-limit page's own examples contradict its table, which costs an agent a turn (12 of 20). clientOid on orders makes retries safe; no MCP annotations to read (10 of 20). Three required fields on a spot order, official SDKs in five languages (11 of 15). 62.