deploy_file_contents is what separates this from a CLI wrapper: files that exist only in the conversation go straight to Cloud Run without being written to disk first. get_service_log sits in the same surface, so a deploy that fails is diagnosed in the same turn it failed.
Google's own Cloud Run MCP server, eight tools: list and create GCP projects, list and inspect Cloud Run services, read their logs, and deploy from a local folder, from file contents passed in the conversation, or from a container image URL.
- Deploy three ways — a local folder by absolute path (deploy_local_folder), files whose contents exist only in the current conversation (deploy_file_contents), or an existing image (deploy_container_image).
- See what is running: list_services for a project, get_service for one service's details.
- Read logs and error messages for a service without leaving the session (get_service_log).
- List the GCP projects available, or create one and attach it to the first available billing account (list_projects, create_project).
A Google Cloud account, with the project and billing you intend to deploy into, and Node.js on the machine. Apache-2.0.
One command plus a key — gemini extensions install https://github.com/GoogleCloudPlatform/cloud-run-mcp, then supply credentials