Anthropic Model Connector
Configure Claude access for Harness Worker Agents with API keys or AWS, GCP, and Azure cloud provider connectors, including IAM roles and OIDC.
The Anthropic Model Connector defines Claude as the LLM provider and default model for your Worker Agents. Authenticate with an Anthropic API key, a Bedrock API key, or an existing AWS, GCP, or Azure cloud provider connector.
If your Claude models are exposed through a LiteLLM OpenAI-compatible endpoint, use the OpenAI Model Connector setup for LiteLLM.
What you will learn in this topic
By the end of this topic, you will be able to:
Before you begin
Before you configure the Anthropic Model Connector, ensure you have the following:
Harness AI enabled: Harness AI must be active for your account. Go to Overview for more information on activating Harness AI.
Model provider access: Access to the Claude models you intend to use through your selected provider. For LiteLLM, use the server, model, and API key prerequisites in the LiteLLM setup.
Authentication credentials: For Personal Token or Bedrock API Key, store the corresponding API key as a Harness text secret. For Cloud Provider, configure an AWS, GCP, or Azure connector with the roles and permissions needed for model access.
Cloud connector access: When you use Cloud Provider, you must have permission to use the referenced connector.
Connector permissions: Permission to create and manage connectors in your Harness project. Go to RBAC in Harness for more information on permissions.
Supported models
The Anthropic Model Connector supports the following Claude models. Select the model that best fits your Worker Agent use case based on complexity, speed, and cost requirements.
Claude Opus 4.8
Latest and most capable model for complex reasoning
Claude Opus 4.7
High-capability model for complex reasoning
Claude Opus 4.6
High-capability model for complex reasoning
Claude Sonnet 4.6
Fast, high-capability model for most tasks
Claude Sonnet 4.5
Previous-generation fast model
Claude Haiku 4.5
Lightweight, low-latency model for simple tasks
Connector configuration
In the connector Details step, select an Authentication method:
Personal Token
Use an Anthropic API key stored as a Harness secret for direct Anthropic access.
Bedrock API Key
Use an Amazon Bedrock API key stored as a Harness secret and configure the region for your model.
Cloud Provider
Reference an existing AWS Connector, GCP Connector, or Azure Connector to obtain credentials through its configured authentication method, including IAM roles or OpenID Connect (OIDC).
After you configure authentication, set the default Model Name. Complete the connectivity setup and connection test, then save the connector.
Cloud provider authentication
Select Cloud Provider to reuse authentication from an existing Harness cloud provider connector. Harness uses that connector to obtain credentials for model access. Configure IAM roles, OIDC federation, or other supported credentials on the cloud provider connector before referencing it from the Anthropic connector.
Use the setup guide for your cloud provider:
AWS Connector
Go to Add an AWS connector to configure IAM role or OIDC authentication.
GCP Connector
Go to Connect to Google Cloud Platform to configure inherited delegate credentials or OIDC authentication.
Azure Connector
Go to Add a Microsoft Azure connector to configure managed identity or OIDC authentication.
The referenced connector must have the cloud permissions required to access your selected model.
To use the cloud provider connector:
In Authentication, select Cloud Provider.
In Select Cloud Provider, choose AWS Connector, GCP Connector, or Azure Connector.
In Connector, select the existing Harness connector for that provider.
In Model Name, select or enter the model identifier for your provider. For example, the AWS configuration below uses
global.anthropic.claude-sonnet-4-6.Select Continue to configure connectivity.
Complete the connection test and save the connector.
Claude through LiteLLM
To use a Claude model configured on a LiteLLM server through its OpenAI-compatible API, create an OpenAI connector that points to the LiteLLM endpoint. LiteLLM routes the requests to the model configured on your server.
Configure the connector with these values:
URL: Your hosted LiteLLM URL with
/v1appended, such ashttps://litellm.example.com/v1.API Key: A Harness secret containing the LiteLLM API key authorized to access the model.
Model Name: The exact model name configured on the LiteLLM server for your Claude model.
Select this connector as the Worker Agent Model Provider. In the agent YAML tab, explicitly set with.backend to openai under agent, even when LiteLLM routes requests to Claude. Note: This is a required manual setting for the OpenAI-compatible LiteLLM flow.
Go to LiteLLM connector configuration to follow the full connector setup, and Use the LiteLLM connector in a Worker Agent to configure the agent.
Runtime model override
Override the connector default model on individual Worker Agents to use different Claude models for specific tasks.
Each Worker Agent uses the default model configured on the connector unless you set the optional Model Name field on the agent. For Anthropic connectors, the Model Name field also accepts an AWS Bedrock inference profile ARN, for example arn:aws:bedrock:us-east-1:123456789012:application-inference-profile/a1b2c3d4e5f6.
Next steps
Worker Agents: Create and configure agents that use this connector.
OpenAI Model Connector: Configure OpenAI as your model provider.
Last updated
Was this helpful?