alpaca-broker-sse-events
Consume Alpaca Broker API real-time event streams over Server-Sent Events (SSE) — account status, journal, transfer/funding, trade, and non-trade-activity events — reliably. Covers connection, auth, replay cursors (since/since_id), heartbeats, reconnection/backoff, ordering, and idempotent processing. Use when building an event consumer for Alpaca lifecycle events in any language.
alpacahq · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add alpacahq/alpaca-skills --skill alpaca-broker-sse-events技能说明
name: alpaca-broker-sse-events description: Consume Alpaca Broker API real-time event streams over Server-Sent Events (SSE) — account status, journal, transfer/funding, trade, and non-trade-activity events — reliably. Covers connection, auth, replay cursors (since/since_id), heartbeats, reconnection/backoff, ordering, and idempotent processing. Use when building an event consumer for Alpaca lifecycle events in any language.
Alpaca Broker API — Real-Time Events (SSE)
Alpaca pushes brokerage lifecycle events over Server-Sent Events: a long-lived HTTP GET that streams text/event-stream. This is not the market-data WebSocket (alpaca-broker-market-data) — different transport, different auth, different reliability model.
Read
alpaca-broker-integrationfirst. SSE uses the Broker API host + HTTP Basic auth (same credential as Broker REST).
Reference
- Guide:
https://docs.alpaca.markets/docs/sse-events - Live:
alpaca-docsMCP →search"SSE Events", thenfetch us/sse-events
1. Why SSE (and why it's simpler than it looks)
SSE is plain HTTP. You don't need a special client: open a GET, keep the connection open, and read the body line-by-line. Each event is a data: line containing a JSON object. It is replayable — you can ask for events from a point in the past and seamlessly catch up to live, which makes it far better than naive polling for lifecycle state.
2. Event streams
| Stream | Path | Carries |
|---|---|---|
| Account status | GET /v1/events/accounts/status | Account-property changes: status/crypto_status (SUBMITTED→ACTIVE, ACTION_REQUIRED, REJECTED), plus kyc_results, account_blocked, trading_blocked, cash_interest, options |
| Journal status | GET /v2/events/journals/status | JNLC/JNLS lifecycle (queued→executed, correct…) |
| Funding/transfer status | GET /v2/events/funding/status | Unified: Transfer, BankRelationship, WireBank, FundingWallet entities (switch on entity_type) |
| Trade updates | GET /v2/events/trades | Order events in the event field: new, fill, partial_fill, canceled, rejected, held, trade_bust, trade_correct… (richer than order status) |
| Non-trade activities | GET /v1/events/nta | Dividends, interest, fees, splits, ACATs, cash disbursements. entry_type e.g. JNLC/FEE/INT/DIVNRA/CSD; status ∈ executed/correct/canceled |
Paths & versions are NOT uniform — verify each. This is exactly the kind of cross-stream inconsistency Alpaca's docs under-communicate:
/v2streams (trades, journals/status, funding/status) use a ULIDevent_iddirectly;/v1/events/tradesand/v1/events/journals/statusare legacy (existing partners only — migrate to v2)./v2/events/tradeswas previously/v2beta1, now redirected./v1streams (accounts/status, nta) are current, not deprecated — there is no v2 yet. Each event carries both an integerevent_idand a ULIDevent_ulid.Every event carries
at(timestamp),account_id, andstatus_from/status_to(account/journal/funding) orevent+order(trades).
3. Connection
GET /v2/events/journals/status?since_id=<last-ulid-you-saw> HTTP/1.1
Host: broker-api.alpaca.markets
Authorization: Basic <base64(key:secret)>
Accept: text/event-stream
Read the response stream and parse data: {…} frames as they arrive. In most languages an off-the-shelf EventSource/SSE client works — just make sure it lets you set the Authorization header on the initial request (the browser EventSource API famously does not; use a server-side SSE library instead).
4. Replay cursors — the feature that prevents data loss
Every stream supports point-in-time replay:
| Param | Meaning |
|---|---|
since / until | Date or RFC3339 timestamps. URL-encode + in offsets as %2B. |
since_id / until_id | ID cursors. On v2 streams the ID is a ULID. On v1 streams it's the integer event_id. |
since_ulid / until_ulid | v1 streams only (accounts, nta) — ULID-based cursors, since v1 events carry both an int event_id and a event_ulid. |
Rules: since is required if until is set; since_id required if until_id set (same for since_ulid/until_ulid); you can't mix since, since_id, and since_ulid. Without any since cursor, no history is returned — you only get live pushes from now on. Reaching the until bound ends the stream with a 200.
This is the single most important reliability lesson: persist the ID of the last event you successfully processed. On every (re)connect, pass it as your since cursor (since_id on v2; since_ulid or since_id on v1) so Alpaca replays anything you missed during the gap. A consumer that reconnects without a cursor silently drops every event that occurred while it was down.
5. Ordering caveat
Within a millisecond, ULIDs contain a random component, so two events in the same millisecond can sort either way. Alpaca's own guidance: for reconciliation, restart the stream from a since a few minutes before your last event and rely on idempotent processing to absorb the overlap. Don't assume strict total ordering — assume approximate ordering plus dedup.
6. Reliability patterns (hard-won)
SSE connections drop — networks, load balancers, deploys, and Alpaca-side resets all happen. A production consumer needs:
- Heartbeat / silence detection. SSE has no application heartbeat by default. Track
lastMessageAton every frame; if the stream is silent past a threshold (e.g. 5 min), proactively tear down and reconnect — a dead socket often looks "open." - Reconnect with exponential backoff + cap. On error/close, reconnect after a delay that doubles up to a ceiling (e.g. start 1s, cap 60s). Reset the delay on a successful connect.
- A single-reconnect guard. Use a flag so an error storm doesn't spawn many concurre