Universal CLI to install, authenticate and execute SaaS tools from terminal. Exposes installed SaaS tool commands as MCP tools via stdio transport.
Single dynamically-registered tool with critical definition gaps. Tool names are generated as '<tool>_<command>' from manifest files, but the MCP server code shows NO explicit tool registration with schemas visible in source. Input schema is 'dynamically built from manifest command.params array', this is inferred, not directly visible. The buildInputSchema() function converts manifest params to Zod, but without seeing actual manifest files, schema quality cannot be verified. Description is constructed at runtime from manifest data ('[tool_name] command_description (HTTP_METHOD path)'), making it impossible to audit for LLM-friendliness. No parameter descriptions are visible in the MCP server code, they come from manifest files external to this codebase. No output schema is documented. Error handling in handler.ts is not shown. This is a meta-tool pattern where actual tool definitions are external and dynamic, making static code analysis impossible. Per HARD SCORING RULES: tool definitions are inferred, not directly visible → cap at 50. Schemas are not visible in source → schema score 0. Descriptions are constructed at runtime from external data → cannot verify LLM quality.
Dynamically registered MCP tools based on installed tool manifests. Each tool is named <tool>_<command> (e.g., stripe_list_customers). The tool description is constructed as: [tool_name] command_description (HTTP_METHOD path)
Tool definitions are external and dynamic, schemas, descriptions, and parameters are loaded from manifest files at runtime, not visible in source code. Cannot audit for LLM-friendliness, parameter types, or output structure.
No input schema visible in MCP server code. buildInputSchema() infers Zod types from manifest params, but manifest structure and actual param descriptions are not shown. Cannot verify if params have descriptions, enums, or constraints.
No output schema documented. Tool descriptions are constructed at runtime as '[tool_name] command_description (HTTP_METHOD path)', no indication of what fields are returned or how to chain tools.
Error handling not visible. handleToolCall() is imported but not shown. Cannot verify if errors guide recovery, categorize as retryable, or include actionable messages.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 24 | 2026-07-28+ | v2 |
Tool naming pattern '<tool>_<command>' is generic and does not start with action verbs (get_, create_, update_, delete_, search_, list_). Actual verb depends on manifest, cannot audit naming quality.