Loading ToolBench
Community-friendly, agentic GitHub security auditing and remediation platform with CLI, MCP, Claude Code, Cursor, reports, PRs, tests, and audit trails.
Server has 5 tools with explicit schemas and descriptions. Naming follows verb_noun convention (list_, security_, dependabot_, remediation_, remediate_). Descriptions are present but vary in quality (10-180 chars). All tools have input schemas with type definitions. Critical gaps: parameter descriptions are minimal or absent (e.g., 'repo' param lacks format guidance); output schemas are undocumented; error handling lacks recovery guidance; no tool annotations (readOnlyHint/destructiveHint); destructive tool (remediate_repository) lacks confirmation pattern despite high risk.
List open Dependabot alerts and their GitHub-provided patched versions.
List active, non-fork repositories visible to the configured GitHub token.
Apply the deterministic security remediation runner. Disabled unless MCP_ALLOW_MUTATIONS=true and execute=true. PR mode is recommended.
Create a conservative remediation plan from open npm Dependabot alerts. This tool never changes a repository.
Return current open Dependabot, code scanning and secret scanning counts for a repository.
Parameter descriptions missing or trivial. 'repo' param across all tools lacks format guidance (e.g., 'Repository name without owner/ prefix, 2-100 chars, alphanumeric + dash/dot/underscore'). LLMs cannot infer constraints from regex alone.
Output schemas undocumented. Tools return JSON via text() wrapper but no schema is declared. LLMs cannot plan downstream calls or extract fields reliably. E.g., dependabot_alerts returns array of {number, package, severity, ...} but this is invisible to the protocol.
Destructive tool (remediate_repository) lacks confirmation pattern. Despite high risk (mutations, PR creation, test execution), tool relies only on execute=false default and MCP_ALLOW_MUTATIONS env var. No dry-run confirmation step or explicit user approval flow.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
No tool annotations. Tools lack readOnlyHint/destructiveHint/idempotentHint metadata. Clients cannot distinguish safe discovery tools from dangerous mutation tools without parsing descriptions.
Error handling lacks recovery guidance. Tools throw generic errors (e.g., 'Repository ... is not available') without suggesting next steps. No categorization of retryable vs user-fixable errors.