Summary

One connection instead of eight consoles.

The value is in the chain rather than any single tool: find the error group, pull the logs around it, jump from a log entry to its trace, then check whether the fix you want to deploy has the IAM permissions to land. Billing sits in the same server, which is unusual — cost anomaly questions get answered in the same session as the incident that caused them.

What it is

An MCP server that connects to Google Cloud and exposes the observability and administration surfaces an on-call engineer reaches for: Logging, Monitoring, Trace, Error Reporting, Profiler, IAM, Spanner and Billing. Several tools accept a natural-language query instead of a filter expression.

What you get

Log entries queried by project, severity, resource type, service and time range — `gcp-logging-query-logs`, `gcp-logging-query-time-range`, `gcp-logging-search-comprehensive`
Metrics retrieved and metric types enumerated, including from a plain-language question — `gcp-monitoring-query-metrics`, `gcp-monitoring-list-metric-types`, `gcp-monitoring-query-natural-language`
Distributed traces fetched, listed, and found from a log entry — `gcp-trace-get-trace`, `gcp-trace-list-traces`, `gcp-trace-find-from-logs`, `gcp-trace-query-natural-language`
Error groups listed with detail and trend analysis — `gcp-error-reporting-list-groups`, `gcp-error-reporting-get-group-details`, `gcp-error-reporting-analyse-trends`
IAM policy read, permissions tested, and deployment permission gaps analysed before you ship — `gcp-iam-get-project-policy`, `gcp-iam-test-project-permissions`, `gcp-iam-validate-deployment-permissions`, `gcp-iam-analyse-permission-gaps`
Spanner instances, databases and tables listed, and SQL executed against them — `gcp-spanner-list-instances`, `gcp-spanner-list-databases`, `gcp-spanner-list-tables`, `gcp-spanner-execute-query`
Billing accounts, services and SKUs read, with cost analysis, anomaly detection and recommendations — `gcp-billing-analyse-costs`, `gcp-billing-detect-anomalies`, `gcp-billing-cost-recommendations`, `gcp-billing-service-breakdown`
CPU and heap profiles listed and compared across deployments — `gcp-profiler-list-profiles`, `gcp-profiler-analyse-performance`, `gcp-profiler-compare-trends`

Requirements

Google Cloud credentials, either as a service account key file pointed at by `GOOGLE_APPLICATION_CREDENTIALS` or as `GOOGLE_CLIENT_EMAIL` and `GOOGLE_PRIVATE_KEY` set directly; `gcloud auth application-default login` produces the former. `GOOGLE_CLOUD_PROJECT` sets the default project, otherwise it is taken from the credentials. Install is a clone plus `pnpm install` and `pnpm build`, then launching `dist/index.js` with node; the package.json names it `google-cloud-mcp` at 0.5.0. If the server times out on startup, `lazyAuth: true` defers authentication until a tool is actually called.

Setup effort

One command plus a key — npx -y google-cloud-mcp, then supply credentials