Troubleshooting
MCP Server Does Not Start
Check local dependencies:
make setup
make run
Check the health endpoint:
curl -fsS http://127.0.0.1:8000/health
DevOps Agent Cannot Connect
Verify:
- endpoint uses HTTPS for AWS DevOps Agent registration
- SigV4 region matches the API Gateway region
- SigV4 service name is
execute-api - Agent invoke role can call
execute-api:Invoke - API Gateway route points to
/mcp
Investigation Returns UNKNOWN
Common causes:
- logs have not landed in S3 or CloudWatch yet
- time window is too narrow
- wrong deploy mode or log layout
- missing
step_id,job_run_id, or application identifiers - Harrier lacks permission to read log buckets or log groups
Use warnings and inconclusive_checks in the diagnosis report to decide what
evidence is missing.
EMR On EKS Has No Pod Evidence
This is recoverable. Harrier can still use EMR Containers, S3, and CloudWatch evidence.
Pod evidence requires one of:
- readable in-cluster Kubernetes config
- readable kubeconfig
kubectlonPATH- namespace-scoped RBAC for pods and events
PR Preview Is Blocked
PR creation is intentionally gated. Dry-run previews should work without write configuration. Real PR creation requires:
- request sets
dry_run=false - request sets
allow_pr_creation=true - server sets
HARRIER_ALLOW_PR_CREATION=true - repo is in
HARRIER_PR_REPO_ALLOWLIST - GitHub token is configured