Skip to content

Common Issues

Reload the AI tool, confirm the MCP config path for that agent, and verify the desktop app is running.

Codex requires /hooks trust for unmanaged command hooks. Review and trust the hook in the CLI.

Add project rules from Rules for CLAUDE.md and AGENTS.md, then ask it to start with list_services.

Credentials Exist But Agent Does Not Log In

Section titled “Credentials Exist But Agent Does Not Log In”

Check that debug_credential_hints are visible, then tell the agent to call get_debug_credentials instead of fabricating tokens.

The local frontend deployment must have a web entry and AI browser debugging enabled. Browser tools only cover local loopback targets.

browser_evaluate is disabled by default. Prefer snapshot, console, network, and screenshot tools; enable evaluate only when needed.

Use list_code_debug_targets first. The target must be a local managed language runtime and already running.

Call list_hosts and use non-local hosts[].id, not host names or IPs.

Check validate_project_pipeline, then inspect list_pipeline_runs, read_pipeline_run_logs, and list_pipeline_artifacts.