Loading ToolBench
MCP server for findarepo.com that finds the best MCP servers, Claude skills, and trending GitHub repos, ranked daily by measured star velocity and commit cadence.
Six tools with clear action verbs and well-structured schemas. Descriptions are detailed and contextual (150-250 chars), explaining WHEN to use each tool. All parameters have types and descriptions. Output is text-based (not structured JSON), which limits downstream chaining. No tool annotations (readOnlyHint, idempotentHint). Error handling is minimal, no recovery guidance or actionable error messages. Schemas are complete but output format is unstructured text rather than typed objects.
Check whether a GitHub repo's stars look organic or suspicious (e.g. sudden spikes, bot-like patterns). Returns findarepo's credibility score and analysis.
Compare two GitHub repositories side by side using findarepo's measured data (stars, recent growth, language, category). Use when a user asks "X vs Y" for tools/libraries.
Get the most consistently developed GitHub repositories, ranked by findarepo's measured weekly commit cadence over the last 26 weeks. This answers "will this project still be maintained?" — a different question from what get_trending_repos (right-now popularity) or star counts (past popularity) answer. Use when a user asks which projects are healthy, alive, or worth depending on long term, or asks for actively maintained alternatives.
Get the fastest-growing GitHub repositories right now, measured by findarepo's daily star-velocity tracking. Optionally filter by category or language. Use for "what's trending on GitHub", "fastest growing AI agent repos", etc.
Search findarepo's directory of Claude skills, plugins, subagents and extensions, ranked by measured GitHub momentum. Use when a user wants to find or recommend a Claude skill/plugin/extension.
Output is unstructured text, not typed JSON objects. Agents cannot extract structured data (repo URLs, star counts, activity scores) for downstream chaining or programmatic use.
No tool annotations (readOnlyHint, idempotentHint, destructiveHint). All tools are read-only but this is not declared, forcing LLMs to infer safety.
Error handling is minimal. No recovery guidance (e.g., 'Unknown category, valid slugs are: ai-agents, web-scraping, ...'). Errors are returned as plain text without actionable next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Search findarepo's directory of MCP (Model Context Protocol) servers, ranked by measured GitHub star velocity. Use when a user wants to find or recommend an MCP server for a task (e.g. "an MCP server for Postgres", "best MCP for web scraping").
Telemetry ping is fire-and-forget and non-blocking, but users are not explicitly informed in the tool description that usage data is collected (even if anonymized). Transparency is missing.
No pagination support in output. Results are capped at limit (max 50) but no next_cursor or total_count returned, preventing agents from discovering if more results exist.