Use Claude Code

POST /v1/messages and POST /v1/messages/count_tokens share the loaded model and admission controller with OpenAI chat. The adapter supports system prompts, text, base64 image blocks, client tools, tool_use/tool_result, thinking output, and Anthropic SSE framing. Server-executed tools are not run by TensorRT Edge-LLM.

Use the recommended sequence length. Raise either limit if Claude Code’s prompt does not fit. Omit --cache-dir to use the default cache. --enable-auto-tool-choice is required because Claude Code sends tool definitions.

edgellm serve Qwen/Qwen3.5-4B \
  --max-input-len 32768 \
  --max-kv-cache-capacity 40960 \
  --enable-auto-tool-choice \
  --served-model-name qwen35-local

ANTHROPIC_BASE_URL=http://localhost:8000 \
ANTHROPIC_API_KEY=local \
ANTHROPIC_DEFAULT_OPUS_MODEL=qwen35-local \
ANTHROPIC_DEFAULT_SONNET_MODEL=qwen35-local \
ANTHROPIC_DEFAULT_HAIKU_MODEL=qwen35-local \
claude

When --api-key is set, both OpenAI bearer authentication and Anthropic x-api-key authentication are accepted.

A one-shot claude -p reply on Qwen3.5-4B was ready. The matching POST /v1/messages response was:

{"content":[{"type":"text","text":"Ready."}],"stop_reason":"end_turn"}