4 views
-/https://github.com/berriai/litellm/issues/26163
GitHub · issue

#26163 [Bug]: `websearch_interception` Anthropic follow-up call passes duplicate Claude Code params (`context_management`, `output_config`) and breaks server-side web search

  • State: open
  • Author: @leike0813
  • Labels: bug, proxy, llm translation, claude code

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

When using Claude Code through LiteLLM's Anthropic `/v1/messages` endpoint with `websearch_interception` enabled, the initial web search interception works, the search provider is called successfully, but the follow-up Anthropic request crashes inside LiteLLM because duplicate keyword arguments are passed to `anthropic_messages.acreate()`.

In my case, the search itself completed successfully, but the agentic follow-up failed with:

1. `got multiple values for keyword argument 'context_management'` 2. after locally hotfixing that, the next failure was `got multiple values for keyword argument 'output_config'`

As a result, Claude Code sees web search as unavailable / unusable even though LiteLLM did perform the search.

## Environment

- LiteLLM: `v1.83.9-nightly` - Release commit: `850fe59` - Claude Code: `2.1.116` - Endpoint used by Claude Code: LiteLLM unified Anthropic endpoint `/v1/messages` - Search provider: SearXNG - Upstream model provider: Anthropic-compatible third-party endpoint - Reproduced on: April 21, 2026

## Relevant config

`…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (1 events)
#0 of 0 · 31d17h39m14s ago — entered · #import:https:::github.com:berriai:litellm post #3211
The left issue is harder because it crosses interception orchestration, request translation, and provider-specific follow-up construction, requiring careful regression coverage for multiple parameter paths. The right issue is comparatively localized to optional-dependency handling and exception classification, with a narrower fix and test surface.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search