MCP stdio server that exposes tool specs as MCP tools. Serves tool actions and management commands (search, inspect, install, run, code) via JSON-RPC over stdin/stdout. In global mode, management tools let agents discover and install new tools dynamically. In tools-only mode, only pre-loaded tool actions are exposed.
Five tools with mixed quality. clictl_search, clictl_list, clictl_inspect have reasonable descriptions (50-150 chars) and clear naming. clictl_install and execute_code lack critical safety documentation. Input schemas are minimal, only 1-2 params per tool with basic type info. No output schemas documented. execute_code is particularly concerning: accepts arbitrary JavaScript with no sandboxing hints, no error recovery guidance, and no permission/scope declaration. Tool names follow verb_noun pattern (search, list, inspect, install, execute) but descriptions lack dependency hints and error guidance. No pagination, no structured output specs, no idempotent markers.
Get tool details including auth requirements and available actions. Show detailed information about a tool including all actions, parameters, auth requirements, and usage examples.
Install a tool by name. Downloads the tool, verifies integrity, and installs to the appropriate location.
List all available tools in the registry, optionally filtered by category.
Search the clictl registry for tools by keyword. Returns matching tool names, categories, and descriptions. Use this BEFORE writing code to call any external API or service.
Execute JavaScript code with access to API clients. Write code that calls the typed functions below. Use console.log() to output results.
execute_code lacks security boundaries and error guidance. Accepts arbitrary JavaScript with no mention of sandboxing, available APIs, timeout limits, or what happens on failure. LLM cannot reason about safety or recovery.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., tool IDs for chaining). Violates pattern:tool and mxe:include-chaining-ids.
clictl_install description does not state it modifies state or warn about side effects. Agents need to know this is irreversible and may require confirmation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Input schemas are minimal. clictl_search and clictl_list accept only 1-2 string params with no enums, ranges, or format constraints. No validation rules in descriptions (e.g., max query length, valid categories).
No error handling guidance. Descriptions do not explain what happens on failure (tool not found, network error, permission denied) or how LLM should recover.