Troubleshooting

"MCP server failed to load" / checks finish in 0 seconds

The bundled MCP server needs uv. Install it:

curl -LsSf https://astral.sh/uv/install.sh | sh

If checks complete instantly and the agent reports that the review couldn't run, the cause is usually missing AWS credentials or an unset region. See the items below.

All checks fail with "Unable to locate credentials"

Configure AWS credentials (run aws configure, set AWS_PROFILE, or export keys) and retry. Verify with aws sts get-caller-identity.

Checks fail asking for a region

Include the region in your request (for example, "review eks-prod in us-east-1") or set AWS_REGION.

"AccessDeniedException" calling Bedrock InvokeModel

Enable model access in the Bedrock console (under Model access) for your region, then confirm your IAM principal has bedrock:InvokeModel.

"No cluster found" / cluster not found

Make sure AWS_REGION is set to the region where the cluster lives, and that your credentials point at the right account. Verify with aws sts get-caller-identity and list clusters with aws eks list-clusters --region <region>. eksreview connects to the Kubernetes API itself using short-lived STS tokens, so you do not need to run aws eks update-kubeconfig. Your IAM identity must still be mapped into the cluster (see Permissions).

Session feels expensive or slow

Type /context to see token usage and cost. Switch to a cheaper model mid-session with /model sonnet, or start a fresh session.