Loading ToolBench
Community Supabase MCP server that enables Cursor and Windsurf to end-to-end manage your Supabase project, execute SQL queries, and more.
This server has critical definition quality gaps. Tool names lack clear action verbs (CALL_AUTH_ADMIN_METHOD, SEND_MANAGEMENT_API_REQUEST are overly generic). Descriptions are present but minimal (10-50 chars), providing insufficient context for LLM tool selection. Input schemas exist but are sparse: CALL_AUTH_ADMIN_METHOD accepts 'method' as a free-form string with no enum constraint, inviting hallucinated method names. EXECUTE_POSTGRESQL accepts raw SQL with no validation hints. Two tools (CONFIRM_DESTRUCTIVE_OPERATION, LIVE_DANGEROUSLY) appear to be safety/confirmation tools but lack proper descriptions and parameter schemas. Output schemas are not documented. Error handling is not visible in the provided code. The server exposes dangerous operations (raw SQL execution, destructive confirmations) without clear permission gates or audit trails.
Calls a method of the Python SDK client for Supabase Auth Admin operations
Confirm a destructive operation before execution
Execute PostgreSQL queries against the Supabase database
Get the specification of available Auth Admin methods
Execute operations without safety checks
Send requests to the Supabase Management API
CALL_AUTH_ADMIN_METHOD accepts 'method' as unconstrained string with no enum. LLM will hallucinate invalid method names (e.g., 'get_user_by_name' when only 'get_user_by_id' exists). Requires enum constraint listing all valid methods.
EXECUTE_POSTGRESQL accepts raw SQL query with no validation, sanitization hints, or injection warnings. No description of SQL dialect, transaction behavior, or result limits. Exposes SQL injection risk if LLM constructs queries from user input.
CONFIRM_DESTRUCTIVE_OPERATION and LIVE_DANGEROUSLY lack input schemas entirely (schema score 0). LIVE_DANGEROUSLY has no description (10 chars). These safety-critical tools are under-specified and cannot guide LLM behavior.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., user_id from create_user response). Violates pattern:tool requirement.
Tool names are generic and lack clear action verbs. 'CALL_AUTH_ADMIN_METHOD' and 'SEND_MANAGEMENT_API_REQUEST' are pass-through wrappers that don't convey intent. Should be split into specific actions: get_user, create_user, list_users, etc.