Loading ToolBench
MCP server for HVTracker trust checks. Check trust signals for AI agents and MCP servers.
HVTracker MCP has 8 well-named, read-only tools with clear descriptions (avg 156 chars, within 10-1024 baseline). All tools start with action verbs (verify, check, compare, search, scan, list, get). Input schemas are present with type declarations and descriptions. However, output schemas are NOT documented in the source, responses are inferred from API calls but not formally declared. Parameter descriptions are adequate but lack format constraints (e.g., no enum for 'category', no min/max for 'limit'). Error handling returns structured fallbacks but lacks recovery guidance. Tool composition is sound (each does one thing), and all are read-only with proper annotations.
Get the HVTracker trust profile for a tracked AI agent or framework, including its runtime capability surface and the URL of its Ed25519-signed trust credential (verifiable offline).
Compare two tracked AI agents side by side: both trust profiles, an evidence-based one-line verdict, and the HVTracker compare-page URL when one is published.
Get the HVTracker trust history for an agent: a timeline of trust score changes, evidence updates, and verdict shifts.
Get the HVTracker leaderboard for a category, ranked by trust score.
List HVTracker categories with agent counts (most-populated first), so you can then pull a category's leaderboard.
Bulk pre-connect trust check for a whole dependency set. Paste a requirements.txt, package.json, MCP client config, or a newline/comma list; each item is returned with a trust verdict plus a stack summary.
Output schemas not documented. Tool responses are inferred from API behavior but not formally declared in tool definitions. LLMs cannot plan downstream calls or extract fields reliably without knowing response structure.
Parameter 'category' in search_agents and get_leaderboard lacks enum constraint. Should enumerate valid categories from list_categories response to prevent hallucinated values.
Numeric parameters 'limit' lack min/max bounds. Should declare 1 - 50 range explicitly in schema and description to prevent LLMs from passing absurd values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2025-06-18+ | v2 |
Search tracked AI agents and frameworks by name, repo, or description.
Pre-connect trust verdict for an MCP server or AI agent. Pass a GitHub owner/repo, GitHub URL, npm/PyPI package, display name, slug, or MCP server URL. Unknown servers return trusted=false because HVTracker has no independent evidence, not because harm is proven.
Error handling returns fallback dicts but lacks recovery guidance. When verify_mcp_server fails, check_agent_trust catches it but does not suggest next steps (e.g., 'Try search_agents() to find similar agents').
list_categories has empty input schema (no parameters). Description should clarify that it returns all categories sorted by population, and that results are not paginated.