Hosted BigQuery MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Scored across 4 tools
execute_action and query_data have virtually identical descriptions — both are labeled 'Step 3 (FINAL) of workflow' with the same required prerequisite steps and the same read-only BigQuery note. An agent cannot tell which one to call or whether they are interchangeable, which is the most critical decision point in this server. prompts_menu and query_contexts are clearly distinct from those two.
execute_action, query_contexts, and query_data follow a verb_noun pattern, but prompts_menu is noun_noun and breaks the pattern. The names are readable but the convention is not uniform, and execute_action/query_data share no naming signal that distinguishes their near-identical purpose.
Four tools is a reasonable, tight scope for a read-only BigQuery workflow server (context discovery, action lookup, prompt catalog, query execution). The count is fine; the problem is that the small set still contains a redundant pair rather than a missing slot.
The described workflow (query_contexts for accounts/action details, then execute) covers the read-only query lifecycle, and the explicit read-only OAuth constraint explains the absence of DDL/DML. However, the duplication between execute_action and query_data means one execution path is redundant while prompt handling is limited to listing/viewing with no clear prompt-execution path.
Execute BigQuery action (Step 3 (FINAL) of workflow).
REQUIRED: Complete these steps first using query_contexts:
request='accounts' - Get workspace_id, brand_id, gcp_project_id AND available_actions
request='actions_details' with actions=['xxxxxx', 'xxxxxx', 'xxxxxx'] - Get detailed body structure for specific action(s) using actions from available_actions
IMPORTANT: Always call step 2 to see the exact body structure before executing.
BigQuery is read-only here (OAuth scope bigquery.readonly): only SELECT queries succeed; DDL/DML/scripts are rejected by Google.
Inputs: body (required). Returns: object
Helper for planning InsightfulPipe BigQuery requests.
Step 1: request='accounts' – Retrieve workspace_id, brand_id, gcp_project_id, other source metadata, AND available_actions (all read-only actions you can call).
Step 2: request='actions_details' with actions=['xxxxxx', 'xxxxxx', 'xxxxxx'] – REQUIRED before querying. Replace with actions from available_actions to get the exact request body, required identifiers, fields, and notes for each action.
Optional: request='available_actions' – Standalone call to list all available actions (already included in accounts response).
Optional: request='brands' with brand_id + workspace_id to pull brand guidelines.
Inputs: request (required), actions, brand_id, platform, +1 more. Returns: object
Query BigQuery data (Step 3 (FINAL) of workflow).
REQUIRED: Complete these steps first using query_contexts:
request='accounts' - Get workspace_id, brand_id, gcp_project_id AND available_actions
request='actions_details' with actions=['xxxxxx', 'xxxxxx', 'xxxxxx'] - Get detailed body structure for specific action(s) using actions from available_actions
IMPORTANT: Always call step 2 to see the exact body structure before querying.
BigQuery is read-only here (OAuth scope bigquery.readonly): only SELECT queries succeed; DDL/DML/scripts are rejected by Google.
Inputs: body (required). Returns: object
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Hosted SQL Server MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Hosted Google Ads MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Hosted PostgreSQL MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Hosted MySQL MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Add one secure layer between your agents and this server.