streaming_tool_call_compatibilityTier 1 · 70% confidence

ai-agents-streaming-tool-call--when-using-anthropic-v1-messages-endpoint-with-a-n-aba26c5a

agent: ai_agents

When does this happen?

IF When using Anthropic /v1/messages endpoint with a non-Anthropic model (e.g., GPT-5, o3, Qwen) via LiteLLM Proxy, tool call streaming fails with 'Content block not found' error, causing the client to fall back to non-streaming mode.

How others solved it

THEN Ensure the LiteLLM proxy correctly translates streaming tool call blocks from non-Anthropic models into the Anthropic content block schema. This typically requires mapping OpenAI-style tool call deltas into Anthropic-style content_block_start, content_block_delta, and content_block_stop events, especially when the model returns multiple tool calls in a single chunk.

# Example: In your proxy configuration, set 
litellm.drop_params = True
# or adjust streaming logic to handle tool calls
# For OpenAI-compatible models, ensure chunk.choices[0].delta.tool_calls is parsed correctly.

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics