Loading ToolBench
Multi-service MCP server hub providing unified access to various integrations including AWS S3, AWS SES, Brave Search, Cloudflare, DALL-E, GitHub, Gmail, Google Drive, Google Maps, Linear, Qdrant, Slack, and Twilio
Scoring was not performed
40 of 48 tools have NO visible input schemas in provided source code. Only brave_web_search, brave_local_search, and generate_image show complete JSON Schema definitions with type and description fields. GitHub, Linear, and Google Drive tools are listed but their schemas cannot be verified, this forces inference and violates the verification principle.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 29 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 42 | - | v1 |
Duplicate tool names across modules cause routing ambiguity and LLM confusion: 'send_email' exists in both aws-ses and gmail contexts; 'list_files' in both s3 and google-drive; 'search_issues' and 'create_issue' in both github and linear. The server must disambiguate or namespace these to prevent collisions.
40 of 48 tools have descriptions that are either missing (20 chars or less) or are bare, single-line sentences that fail to explain WHEN to call the tool vs similar alternatives. E.g., 'Create a new issue in Linear' does not distinguish this from GitHub's create_issue.
No output schemas are documented for any tool. LLMs cannot plan downstream calls or extract needed fields without knowing response structure. E.g., does create_pull_request return pull_request_id? Does list_issues return issue_ids, titles, and statuses? Every response field is undocumented.
Parameter descriptions are absent or minimal for nearly all tools except Brave Search. GitHub tools list parameter names but no type, constraint, or usage information. E.g., create_branch has no description of what 'ref' parameter should contain or whether it's branch name or commit SHA.
Error handling is minimal. Brave Search tools show basic try/catch but provide no recovery guidance or error categorization. Other tools have no visible error handling in source code.
No pagination parameters visible for list tools. list_files, list_commits, list_issues, list_pull_requests, etc. show no limit, offset, or cursor parameters in source code. Without pagination, large result sets waste tokens and risk context window exhaustion.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) visible in source. This means agents cannot distinguish read-only tools from destructive ones at a glance. Merge pull request, delete operations, and send_email lack explicit 'destructive' warnings.