Clarify auth tiers: keyless reads work without x-api-key - #20
Conversation
Most read endpoints (/lend/v1/earn/tokens, /prediction/v1/events,
/markets/{id}, /orderbook/{id}) work keyless at 0.5 RPS during
prototyping. Only write endpoints and higher rate limits need an
authenticated key. Replace the blanket "required" claim with a
tiered description so agents reading SKILL.md don't fail-fast on
missing keys for read-only exploration.
Verified against the live API on 2026-05-01 during a hackathon
build. Aligns SKILL.md with the convention already in
https://dev.jup.ag/docs/llms.txt.
|
|
|
||
| **Auth**: | ||
| - **Read endpoints** (e.g. `/lend/v1/earn/tokens`, `/prediction/v1/events`, `/prediction/v1/markets/{id}`, `/prediction/v1/orderbook/{id}`): keyless at 0.5 RPS works — useful for prototyping and discovery. | ||
| - **Write endpoints** (e.g. `POST /prediction/v1/orders`, swap quote/build, transaction crafting) and **higher rate limits**: require `x-api-key` from [portal.jup.ag](https://portal.jup.ag/). |
There was a problem hiding this comment.
https://developers.jup.ag/docs/portal/rate-limits#overview - rate limits are always 0.5 it does not depend on endpoint
4 participants
What
Replace the blanket "auth required for Jupiter REST endpoints" sentence in
integrating-jupiter/SKILL.mdwith a tiered description that distinguishes keyless reads from authenticated writes.Why
SKILL.md content is auto-loaded into agent context (Claude Code, Cursor, etc.) and read as a runtime instruction. The current line:
overstates the requirement. Keyless 0.5 RPS works on the read endpoints we touched while building a hackathon project on Jupiter Prediction + Lend Earn. Confirmed empirically:
GET /lend/v1/earn/tokens→ 200, returned 5 tokens (USDC, USDT, SOL, EURC, JUICED)GET /prediction/v1/events?category=crypto→ 200, returned 4054 eventsGET /prediction/v1/markets/{id}→ 200GET /prediction/v1/orderbook/{id}→ 200The convention in https://dev.jup.ag/docs/llms.txt already documents keyless access correctly — this brings the SKILL.md into alignment with that source of truth.
Effect
Agents reading the SKILL.md verbatim during prototyping no longer fail-fast on missing API keys for read-only exploration. Authenticated tiers remain accurately described for write paths.
Context
This came out of a Solana Frontier hackathon project (Jupiter sidetrack). Full DX report: https://github.com/criptocbas/Ballast/blob/main/DX-REPORT.md — the fix here is captured as
DX-GAP-#13in that report, alongside 26 other findings.Happy to iterate on phrasing or formatting (e.g. a table instead of bullets, additional endpoints listed) if a different style fits the repo's conventions better.