Loading ToolBench
Local MCP server for fetching Reddit content
Single tool 'fetch_reddit' has a clear verb-noun name and reasonable description (118 chars, within 10-1024 range). Input schema is present with proper Zod validation (url as string with format, comment_limit as number with min/max). However, output schema is completely undocumented, the tool returns formatted text but LLMs cannot see what structure to expect. Parameter descriptions are adequate but generic. No error recovery guidance provided to LLMs. Tool is read-only (good), but lacks idempotent/readonly annotations. Missing output documentation is a critical gap for agent composition.
Fetch content from a Reddit URL. Supports post pages (title, body, top comments), subreddit listings, and user pages.
Output schema completely undocumented. Tool returns formatted text but LLMs cannot see expected structure, field names, or how to chain results to downstream tools.
No error recovery guidance. When Reddit returns 429 (rate limit) or 403 (forbidden), error messages are user-facing but do not guide LLM on retry strategy or alternative actions.
Missing tool annotations. Tool is read-only but lacks readOnlyHint annotation in schema. Agents cannot infer safety properties without explicit metadata.
Parameter description for 'url' is generic ('Reddit URL (post, subreddit, or user page)'). Does not explain format constraints, valid domain patterns, or what happens if URL is invalid.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
No pagination or result limits documented. Tool silently truncates comments and replies (MAX_REPLIES_PER_COMMENT=5, MAX_BODY_LENGTH=20000) but LLM is not informed of these limits or how to request more data.