GCP Bucket MCP Server

Read-only MCP server exposing GCPBucketClient to AI agents: list objects in a Google Cloud Storage bucket.

Tools

Tool	Description
`list_objects(prefix: str \| None = None) -> list[str]`	List object names in the configured bucket, optionally filtered by prefix
`read_object_text(remote_path: str, max_bytes: int = 1_000_000) -> str`	Read an object's content as UTF-8 text. Refuses (raises) objects larger than `max_bytes` rather than truncating.
`get_signed_url(remote_path: str, expires_in_seconds: int = 300) -> str`	Time-limited, read-only URL — for content beyond `read_object_text`'s inline cap. Requires `GOOGLE_APPLICATION_CREDENTIALS` to point at a real service account key file (signing needs a private key, not metadata-server ADC).

upload()/download() are write/local-filesystem-shaped and aren't exposed.

The client is cached at module scope rather than rebuilt per call — GCPBucketClient's constructor makes a real network call (an SSL capability probe) plus Application Default Credentials resolution, and neither changes within a running server process.

Tested against a real fake-gcs-server container, not mocks.

Environment Variables

Variable	Required	Notes
`GCP_BUCKET_NAME`	yes	GCS bucket name
`GOOGLE_APPLICATION_CREDENTIALS`	yes	Path to a service account JSON key (read-only storage role recommended). Read by the underlying `google-auth` library via Application Default Credentials, not by this server directly.

Run locally

pip install "data-connectors-ai[gcp,mcp]"
data-connectors-gcp-bucket-mcp

Use with an MCP client

{
  "mcpServers": {
    "gcp-bucket": {
      "command": "uvx",
      "args": ["--from", "data-connectors-ai[gcp,mcp]", "data-connectors-gcp-bucket-mcp"],
      "env": {
        "GCP_BUCKET_NAME": "your-bucket",
        "GOOGLE_APPLICATION_CREDENTIALS": "/path/to/service-account.json"
      }
    }
  }
}

Registry

Published at io.github.nagarjunr/data-connectors-gcp-bucket on the official MCP registry.

GCP Bucket Connector (data-connectors-ai)

Documentation