Claude Code gateway compatibility guide changedllm-gateway-protocol
Upstream edited this page at 5 Oct 2026 01:00 UTC, give or take a minute or two: the time comes from Anthropic’s own sitemap rather than from a commit. This site recorded the change at 5 Oct 2026 01:07 UTC.
Upstream edited
Recorded here
Lines+7added
Lines−9removed
From line
59
where the diff opens
First seen
14 Aug 2026
this site's first read of the page
Recorded edits31to this page, all time
The whole hunk
from line 59, old and new numbered
/
from line 59
5959
6060### Streaming
6161
62Stream inference responses. Claude Code reads the stream as it arrives, so if your gateway buffers complete responses before relaying them, Claude Code stalls.
62Claude Code reads each streaming inference response event by event as it arrives, so the way your gateway relays the stream affects what the user sees:
6363
64Deliver each response's full event sequence without dropping, duplicating, or reordering events. When an Amazon Bedrock guardrail blocks a reply, forward the events it sends unchanged, even when they reference a content block whose `content_block_stop` already arrived. [AWS Guardrails](/docs/en/amazon-bedrock#aws-guardrails) describes how that reply ends. When any other event references a content block whose `content_block_start` never arrived, or a block whose `content_block_stop` already arrived, Claude Code stops reading the stream at that event instead of applying it, so a duplicated `content_block_stop` can't run the same tool call twice. [The response above may be incomplete](/docs/en/errors#the-response-above-may-be-incomplete) describes what the user sees, under the `Part of the response never arrived` and `The response stream was malformed` variants.
65
66Relay each response through its final `message_delta` and `message_stop` events before ending the body. A body that ends after a `message_delta` carrying a `stop_reason`, with no content block still open and no content block event after that frame, counts as complete even when `message_stop` is missing. A body that your gateway ends cleanly any earlier, once a content block has started, is treated the same as a dropped connection: [Automatic retries](/docs/en/errors#automatic-retries) says when Claude Code re-issues the request, and [The response above may be incomplete](/docs/en/errors#the-response-above-may-be-incomplete) covers what it keeps once visible content has arrived. Claude Code keeps the `stop_reason` a `message_delta` delivers, so a later usage-only `message_delta` whose `delta` has `stop_reason: null` or no `stop_reason` key doesn't clear it.
67
68When the client speaks the Amazon Bedrock format, relay the `InvokeModelWithResponseStream` response body and its `Content-Type: application/vnd.amazon.eventstream` header unmodified, and don't convert the stream to server-sent events. See [Streaming errors behind a gateway or proxy](/docs/en/amazon-bedrock#streaming-errors-behind-a-gateway-or-proxy).
69
70Forward keep-alive pings as well, because Claude Code aborts a streaming response once no bytes reach it for [five minutes by default](/docs/en/network-config#streaming-idle-watchdogs). During a long thinking pause, the upstream's SSE `ping` events can be the only bytes on the stream. If your gateway strips or buffers them, Claude Code aborts the response partway through the pause. When you translate from an upstream that sends no pings at all, such as Amazon Bedrock's binary event-stream, emit your own `ping` events during silent gaps.
64* If your gateway buffers a response until it is complete, Claude Code stalls.
65* Claude Code expects each response's full event sequence, in order, through the final `message_delta` and `message_stop` events. If the body ends cleanly after a content block has started but before that final `message_delta`, Claude Code treats the response as a dropped connection. [The response above may be incomplete](/docs/en/errors#the-response-above-may-be-incomplete) describes what the user sees then, and [Automatic retries](/docs/en/errors#automatic-retries) says when Claude Code re-issues the request instead.
66* When an Amazon Bedrock guardrail blocks a reply, the events Bedrock sends can reference a content block whose `content_block_stop` already arrived, and Claude Code relies on receiving them as sent. [AWS Guardrails](/docs/en/amazon-bedrock#aws-guardrails) describes how that reply ends.
67* Claude Code aborts a streaming response once no bytes reach it for longer than its [streaming idle timeout](/docs/en/network-config#streaming-idle-watchdogs). During a long thinking pause, the upstream's SSE `ping` events can be the only bytes on the stream, so a gateway that strips or buffers them can trip that timeout partway through a response. A gateway that translates from an upstream without pings, such as Amazon Bedrock's binary event stream, has the same gap unless it emits its own `ping` events.
68* On the [Amazon Bedrock InvokeModel format](#api-formats), Claude Code reads the `/model/{model}/invoke-with-response-stream` response as the binary `application/vnd.amazon.eventstream` body Bedrock returns, and can't parse it once a gateway converts it to server-sent events or rewrites that `Content-Type` header. [Streaming errors behind a gateway or proxy](/docs/en/amazon-bedrock#streaming-errors-behind-a-gateway-or-proxy) describes what the user sees then.
7169
7270### Format mismatch with the upstream
7371
No line in this hunk matches that.