Loading ToolBench
Dependency-free launch hygiene tools for AI agent and MCP builders.
This is a specialized security audit toolkit with 8 tools, all read-only. Tool names are clear and action-oriented (scan, audit, check, matrix). Descriptions are detailed (150-250 chars) and explain purpose well. Input schemas are present with proper types and enums. However, output schemas are not documented, LLMs cannot predict what fields to expect from audit reports. Error handling is minimal; tools do not guide recovery or classify errors. No tool annotations (readOnlyHint, etc.) despite all being read-only. Parameter descriptions are good but lack examples of valid file paths or format expectations. The tools are narrowly focused (Supabase, MCP, security) and do not compose with other tools, this is intentional for a standalone audit suite, but limits agent integration.
Reads a redacted MCP client config and prints a pre-install risk report. Detects shell wrappers, package runners, unpinned packages, risky shell tokens, broad filesystem paths, non-HTTPS remote URLs, and sensitive environment variables.
Reads an MCP tools/list JSON result and prints a non-sensitive permission matrix. Classifies tools by action class (credential_or_permission, financial_or_identity, execute_code, destructive_write, external_send_or_publish, write_record, read_private, read_public) and gate (allow, ask, deny). Detects metadata injection patterns, schema quality issues, and $ref conformance.
Generates a non-sensitive MCP trust verification plan with checklist criteria for server identity, manifest review, action classification, approval gates, prompt-injection fixtures, and redacted evidence.
Scans only public http(s) URLs you own or have permission to test. Checks for missing security headers (CSP, X-Content-Type-Options, X-Frame-Options, Referrer-Policy, Permissions-Policy), CORS misconfiguration, cookie security flags, and source-map exposure.
Reads redacted Supabase RLS policy SQL and prints an anonymous sign-in authorization report. Detects policies that allow anonymous users to read, write, or accept invitations without explicit is_anonymous guards.
Output schemas not documented. LLMs cannot predict audit report structure (fields, types, nesting). Agents cannot extract specific findings or chain results to downstream tools.
No tool annotations despite all tools being read-only. Adding readOnlyHint=true would signal to agents that these tools are safe to call speculatively without side effects.
Error handling absent. Tools do not return actionable error messages or recovery guidance. A file-not-found error should suggest valid paths or formats; a schema validation error should show what was invalid.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2025-06-18+ | v2 |
Reads redacted Supabase migration SQL and prints a grants cutover audit report. Detects missing GRANT statements, overly broad grants (GRANT ALL), and RLS policy gaps during the 2026 Data API explicit-grants cutover.
Reads redacted Supabase RPC/function SQL and prints a security audit report. Detects SECURITY DEFINER vs SECURITY INVOKER, search_path mutability, default EXECUTE exposure, and function-level access control issues.
Reads redacted multi-tenant Supabase SQL and prints a tenant-boundary audit report. Detects RLS policies that fail to enforce organization/tenant isolation, cross-tenant data leaks, and missing membership checks.
Parameter descriptions lack format hints. 'file' parameters do not specify expected file extensions (.sql, .json, .txt) or whether paths are relative or absolute. 'format' enum is present but no guidance on when to use json vs markdown.
No composition with other tools. These are standalone audit utilities; they do not return IDs or references that downstream tools accept. This is acceptable for a specialized toolkit but limits agent integration.