AWS Bedrock AgentCore

Deploy a harness agent from the catalog to AWS Bedrock AgentCore.

Before you begin

Set up an OIDC provider so that users can authenticate with the registry.
Install Solo Enterprise for agentregistry in an AWS EKS cluster and the arctl CLI.
Port-forward the Solo Enterprise for agentregistry server on port 12121 if you have not done so already. The registry is not exposed externally by default, so you need port-forwarding to reach it from your local machine.
```
kubectl -n agentregistry-system port-forward svc/agentregistry-enterprise-server 12121:12121
```

Connect to the AWS Bedrock AgentCore runtime.
Set up the harness runner image on the Runtime so that the agent can be deployed into the harness container in AWS Bedrock AgentCore.
Create and publish a Model in the catalog so the harness runner knows which Bedrock model to use.
Create a harness agent and publish it to the catalog.

Deploy the agent

List the runtimes that are connected to Solo Enterprise for agentregistry and note the name of the AgentCore runtime you want to deploy to.
```
arctl get runtimes
```
Example output:
```
NAME                 TYPE
agentcore            BedrockAgentCore
kubernetes-default   kubernetes
local                local
```

Create a Deployment for the harness agent. The registry deploys a pre-built harness runner container to AgentCore that is defined in your runtime. At startup, the runner pulls the plugin bundles, assembles the agent, and starts the Claude Code binary in headless mode.

arctl apply -f - <<EOF
apiVersion: ar.dev/v1alpha1
kind: Deployment
metadata:
  name: hello-agent
spec:
  targetRef:
    kind: Agent
    name: hello-agent
    tag: latest
  runtimeRef:
    kind: Runtime
    name: agentcore
  modelRef:
    name: claude-sonnet-4
  harness:
    type: claude-code
    permissionMode: bypassPermissions
  env:
    AGENT_TRACE_CONTENT: "true"
EOF

Field	Description
`spec.targetRef`	The harness agent to deploy. `kind` must be `Agent`; `name` and `tag` must match a published agent.
`spec.runtimeRef`	The runtime to deploy to. Use `kind: Runtime` and the name from `arctl get runtimes`.
`spec.modelRef.name`	The model the harness should use. The registry passes this to Claude Code as the `--model` flag. Defaults to the organization’s default model if omitted.
`spec.harness.type`	The harness to run the agent on. Supported values: `claude-code` and `openclaw`.
`spec.harness.permissionMode`	The Claude Code permission mode. Use `bypassPermissions` for headless deployments with no user in the loop, or `acceptEdits` to allow file edits without further confirmation.
`spec.env`	By default, harness agent tracing spans include structural attributes only, such as `gen_ai.response.model` and token usage. To also capture the invocation’s prompt and completion text on the span, set `AGENT_TRACE_CONTENT` to `true` on the harness agent Deployment.

Example output:

✓ Deployment/hello-agent created

Open the Instances view in the UI and select the agent you deployed. Monitor the deployment progress in the Instance Logs section. The deployment can take a few minutes to complete.

Chat with the agent

After the deployment is complete, chat with the agent through the Solo Enterprise for agentregistry UI.

Open the Instances view, select the hello-agent deployment, and click Open Chat.
Check that the prompt is loaded correctly by asking the agent what its role is.
```
what is your role?
```
Verify that the agent shows the system prompt that you created earlier.
Check that the plugin was loaded correctly by calling the greeting skill in the plugin.
```
hello-plugin:hello testuser
```
Verify that you see the Hello testuser! message with available slash commands to use.
Check that the skill was loaded correctly by asking the agent to echo a message back to you.
```
Use the echo skill to repeat the following message: This is the echoed message from my loaded skill.
```
Verify that the message is echoed back to you.
Check that you have access to the tools of the Solo.io documentation MCP server.
```
List the Solo MCP server tools
```
Verify that you get back the MCP server tools, such as the search tool.
Check that you have access to the built-in Claude Code skills. In addition to any skills you add to the Agent definition, the harness runner always includes Claude Code’s built-in skills, such as /deep-research, /code-review, and /security-review. These are part of the Claude Code CLI itself and are available in every harness agent session.
```
list all the skills that you have
```
Verify that you see default Claude Code skills alongside the skills that you added to the Agent definition.

Next steps

Runtime authorization for AgentCore Control which agents can invoke MCP server tools and call other agents at runtime using AccessPolicies enforced by agentgateway.