langchain - 💡(How to fix) Fix feat(anthropic): expose response headers in response_metadata [1 participants]

Submission checklist

• This is a feature request, not a bug report or usage question.
• I added a clear and descriptive title that summarizes the feature request.
• I used the GitHub search to find a similar feature request and didn't find it.
• I checked the LangChain documentation and API reference to see if this feature already exists.
• This is not related to the langchain-community package.

Package (Required)

• langchain
• langchain-openai
• langchain-anthropic
• langchain-classic
• langchain-core
• langchain-model-profiles
• langchain-tests
• langchain-text-splitters
• langchain-chroma
• langchain-deepseek
• langchain-exa
• langchain-fireworks
• langchain-groq
• langchain-huggingface
• langchain-mistralai
• langchain-nomic
• langchain-ollama
• langchain-openrouter
• langchain-perplexity
• langchain-qdrant
• langchain-xai
• Other / not sure / general

Feature Description

Expose the HTTP response headers returned by Anthropic's API (request-id, anthropic-ratelimit-*, etc.) on AIMessage.response_metadata, behind an opt-in flag — mirroring the include_response_headers parameter already available on ChatOpenAI.

Apply to both streaming and non-streaming paths (invoke, ainvoke, stream, astream).

Use Case

We log every LLM call in our production backend and want to store Anthropic's request-id (req_011...) alongside each call so we can jump from a log line straight to the Anthropic console when debugging. Same story for the rate-limit headers — we want to surface them to ops dashboards.

Right now that header never reaches user code from ChatAnthropic, so we can't correlate application logs with the Anthropic console without resorting to hacks (see Alternatives).

Proposed Solution

Add an include_response_headers: bool = False option to ChatAnthropic, aligned with how it works in ChatOpenAI. When enabled, the response headers should be exposed via response_metadata["headers"].

Implementation would follow the Anthropic SDK's .with_raw_response() path and be applied consistently across _generate, _agenerate, _stream, and _astream.

There was prior work in PR #31579 that implemented this end-to-end (including tests), but it was closed due to inactivity. I can take it over—rebase onto master, address any outstanding review comments, and get it merged.

A few small adjustments I'd make vs. the original PR:

• Streaming: attach headers to the first chunk only (matches ChatOpenAI's is_first_chunk pattern) rather than every chunk.
• response_metadata: add a "headers" key instead of replacing the existing dict, so model_provider and other keys remain untouched when the flag is on.
• Tests: assert on request-id specifically — the original check passed as long as content-type was present, which it always is.

Alternatives Considered

We currently capture the header via an httpx AsyncClient event_hook that stashes the request-id in a ContextVar per request. It works but:

• Ties us to httpx internals (fragile across anthropic-sdk-python versions)
• Doesn't survive if the SDK switches transports
• Every app re-implementing the workaround reinvents the same wheel

langchain-aws solved the equivalent problem officially in #473 (ChatBedrockConverse, released in 0.2.23). Would be good to have parity on the Anthropic side.

Additional Context

• Previous attempt: #31579
• Reference for similar feature in another provider: langchain-ai/langchain-aws#473

Would appreciate being assigned if there's no objection to resuming this. I'll base the PR on current master rather than reviving the stale branch.