Loading ToolBench
Remote MCP server exposing the Sigistry plugin catalog to AI agents (read-only, Streamable HTTP). Provides access to verified Claude Code plugins, skills, and security audit results.
Strong foundation with well-structured schemas, clear descriptions, and proper tool annotations. All 8 tools have explicit registrations with input/output schemas and descriptions. Tool names follow verb_noun convention (search_*, get_*, list_*). Descriptions are detailed and contextual (avg ~180 chars). Input schemas use Zod with type constraints and descriptions. Output schemas are documented with structured Zod objects. READ_ONLY annotations applied consistently. Key gaps: verify_plugin lacks visible implementation (capped at 50); some parameter descriptions could be more prescriptive about constraints; no explicit error recovery guidance in descriptions.
Get the full details of a single Sigistry plugin by its id, including install commands, component counts, and its security-audit result (per-check pass/fail from the Verified by Sigistry methodology).
Get one Sigistry scorecard by id, including the full per-axis results and hardening notes.
Get one Sigistry skill by name, including the full raw SKILL.md source. The source is portable: it can be applied directly in any SKILL.md-aware agent (Claude Code, Claude Desktop, and others) without installing anything, or installed natively in Claude Code via the parent plugin, which keeps it verified and updated. The returned skill passed the Sigistry skill-safety check at the parent plugin's verification date.
List the distinct plugin categories in the Sigistry marketplace with a count of plugins in each, plus the total plugin count.
List independent security grades of public MCP servers from the Sigistry scorecard rubric.
verify_plugin lacks visible implementation and schema. Tool mentioned in comments but no explicit registration or input/output schema shown in source. Cannot verify parameter types or output structure.
Error handling descriptions missing from tool descriptions. Tools return error objects (e.g., 'not_found' for get_plugin) but descriptions do not explain error cases or recovery paths. LLMs cannot infer what to do on failure.
Parameter descriptions lack explicit constraints. 'query' and 'plugin' parameters accept free-form strings with no length limits, pattern constraints, or validation rules stated. LLMs may pass invalid values.
Output pagination not documented. search_plugins returns 'up to 15 matching plugins' but no limit parameter, offset/cursor, or total_count field visible in schema. Large result sets could exhaust context.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 67 | 2026-07-28+ | v2 |
Search the Sigistry marketplace of Claude Code plugins by keyword and/or category. Returns matches with their install command and verification status (the registry runs an eight-check security audit; prefer "verified" plugins when recommending an install).
Search the Sigistry catalog of verified skills (SKILL.md instruction sets for AI agents) by keyword, optionally filtered to one parent plugin. Every skill passed the skill-safety check: no command shadowing, honestly-scoped triggers, no injection or concealment language, no unsafe scripts. Use get_skill to fetch the full portable source of a match.
Local recipe tool for verifying a plugin (mentioned in comments but implementation not shown in truncated source).
Verification status enum not enforced in input. search_plugins and search_skills return verification status as string ('verified', 'stale', 'listed', 'failed', 'unknown') but no enum constraint on input parameters to filter by status.