Prerequisites

• I've searched the current open issues
• I've updated to the latest version of Toolbox

Toolbox version

0.14.0

Environment

Environment

• ✅ Working: Local development → MCP Toolbox
• ❌ Failing: Cloud Run → MCP Toolbox
• ✅ Working: All other BigQuery tools in both environments
• ❌ Failing: Only bigquery-conversational-analytics tool

Linux johanesa45 6.12.32-1rodete1-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.12.32-1rodete1 (2025-06-26) x86_64 GNU/Linux

Client

Client

Via ADK deployed on Cloud Run.

Expected Behavior

Expected Behavioud

• Working on all env.

Current Behavior

Problem

The bigquery-conversational-analytics tool fails with ACCESS_TOKEN_SCOPE_INSUFFICIENT error in Cloud Run environments, while all other BigQuery tools work fine.

Error Details

{
  "error": {
    "code": 403,
    "message": "Request had insufficient authentication scopes.",
    "status": "PERMISSION_DENIED",
    "details": [{
      "@type": "type.googleapis.com/google.rpc.ErrorInfo",
      "reason": "ACCESS_TOKEN_SCOPE_INSUFFICIENT",
      "domain": "googleapis.com",
      "metadata": {
        "method": "google.cloud.geminidataanalytics.v1alpha.DataChatService.Chat",
        "service": "geminidataanalytics.googleapis.com"
      }
    }]
  }
}

Root Cause

The conversational analytics tool uses TokenSource with limited BigQuery scope (bigqueryapi.Scope), but the Gemini Data Analytics API requires broader cloud-platform scope.

Key difference from working tools:

• Other BigQuery tools use ClientCreator pattern for flexible scope handling
• Conversational analytics tool makes direct HTTP calls to geminidataanalytics.googleapis.com with insufficient scopes

Steps to reproduce?

Reproduction

Use this test script to reproduce the error:

#!/usr/bin/env python3
import os
import json
import requests
import google.auth
from google.auth.transport.requests import Request

def test_conversational_analytics():
    # Get cloud-platform scoped token
    credentials, _ = google.auth.default(scopes=["https://www.googleapis.com/auth/cloud-platform"])
    request = Request()
    credentials.refresh(request)

    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {credentials.token}"
    }

    payload = {
        "jsonrpc": "2.0",
        "id": 1,
        "method": "tools/call",
        "params": {
            "name": "bigquery-conversational-analytics",
            "arguments": {
                "user_query_with_context": "What is the structure of this table?",
                "table_references": json.dumps([{
                    "projectId": "your-project",
                    "datasetId": "samples",
                    "tableId": "shakespeare"
                }])
            }
        }
    }

    response = requests.post("YOUR_TOOLBOX_URL/mcp", headers=headers, json=payload)
    print(f"Status: {response.status_code}")
    print(response.text)

if __name__ == "__main__":
    test_conversational_analytics()

Additional Details

Files Affected

• internal/tools/bigquery/bigqueryconversationalanalytics/bigqueryconversationalanalytics.go

Impact

This issue prevents the conversational analytics tool from working in production Cloud Run environments where other BigQuery tools work perfectly. The tool is the only BigQuery tool that exhibits this authentication scope limitation.