Anthropic Messages

Connect AgentCompass to Anthropic Messages-compatible endpoints.

Use the anthropic protocol for Claude-style Anthropic Messages endpoints, commonly exposed at /v1/messages.

Configure the Protocol

export MODEL_NAME=""
export MODEL_BASE_URL="https://your-endpoint.example"
export MODEL_API_KEY="sk-ant-..."

agentcompass run <benchmark> <harness> "$MODEL_NAME" \
  --env <environment> \
  --model-base-url "$MODEL_BASE_URL" \
  --model-api-key "$MODEL_API_KEY" \
  --model-api-protocol anthropic \
  --model-params '{"max_tokens":4096,"temperature":0}'

The CLI protocol id is anthropic, while the documentation and user interface describe it as Anthropic Messages.

Message Conversion

When a runtime component uses the shared Anthropic client, AgentCompass combines chat-style system messages into the Anthropic request’s top-level system field and forwards the remaining messages to the Messages API.

Anthropic endpoints commonly require an explicit max_tokens value. Other request parameters are forwarded when supported by the endpoint and selected harness.

Harness Compatibility

Protocol support is declared by each harness. For example, Claude Code uses Anthropic-compatible credentials, while many other harnesses are designed around OpenAI-compatible protocols. Always confirm compatibility on the selected Harness page.

See Model Configuration for shared fields and OpenAI Chat or OpenAI Responses for the other built-in protocols.