davidcarma/gcp-integration

GCP integration (MCP)

The gcp MCP server uses Application Default Credentials to call GCP REST APIs. It does not open the Cloud Console UI.

This integration is multi-project. It is not tied to one single project. Use gcp_list_projects to discover everything the authenticated account can access, then target the desired projectId explicitly in each call.

User-facing naming rule:

• Prefer the project's friendly display name when talking to the user.
• Keep raw projectId values internal unless they are needed for commands, API paths, debugging, or the user explicitly asks for them.

Prerequisites

• Auth: Run once:

  gcloud auth application-default login
  

• Enable the relevant API in the target project (e.g. Vertex AI API, Compute Engine API).

Tools

| Tool | Purpose | |------|--------| | gcp_list_projects | List projects (optional parent: organizations/ID or folders/ID). | | gcp_get_project | Get one project by project ID. | | gcp_list_organizations | List organizations you can access. | | gcp_request | Generic: call any GCP REST endpoint (method, base_url, path, query, body). |

Generic gcp_request (full GCP surface)

Use this to call Vertex AI, Compute, Storage, IAM, Cloud Run, etc. The agent supplies the REST path and body from GCP docs.

Project targeting rule:

• For multi-project safety, always pass the intended projectId in the request path, for example /v1/projects/PROJECT_ID/....

• Do not assume a hidden default project unless the user explicitly asks to set one.

• When presenting project choices, show friendly display names first and include raw IDs only if needed for disambiguation.

• base_url: API base, e.g. https://aiplatform.googleapis.com, https://compute.googleapis.com, https://storage.googleapis.com, https://cloudresourcemanager.googleapis.com.

• path: Path only, e.g. /v1/projects/PROJECT_ID/locations/us-central1/endpoints.

• method: GET, POST, PATCH, PUT, DELETE.

• query: Optional dict for query string (e.g. {"pageSize": "10"}).

• body: Optional dict for JSON body (POST/PATCH).

Common base URLs

| API | base_url | |-----|----------| | Vertex AI | https://aiplatform.googleapis.com | | Compute Engine | https://compute.googleapis.com | | Cloud Storage (JSON) | https://storage.googleapis.com | | Resource Manager | https://cloudresourcemanager.googleapis.com | | IAM | https://iam.googleapis.com | | Cloud Run | https://run.googleapis.com | | Cloud Functions | https://cloudfunctions.googleapis.com |

Paths and request shapes are in GCP REST API reference and each product’s docs (e.g. Vertex AI REST).

Resource name format

• Organization: organizations/123456789
• Folder: folders/987654321
• Project: use projectId (string like my-project-id).

Examples

• List projects: gcp_list_projects.
• Get one project directly: gcp_get_project with project_id.
• Vertex AI list endpoints: gcp_request with base_url https://aiplatform.googleapis.com, path /v1/projects/PROJECT_ID/locations/us-central1/endpoints, method GET.
• Compute list instances: base_url https://compute.googleapis.com, path /compute/v1/projects/PROJECT_ID/zones/ZONE/instances, method GET.