Loading ToolBench
An AI Employee system that monitors inboxes and action folders, processes tasks via Claude/Qwen, and manages workflow through an Obsidian vault. Includes file system watchers, orchestration, and integration with Playwright MCP for browser automation.
This server exhibits severe definition quality gaps across nearly all dimensions. Tool descriptions are minimal or absent (most under 20 chars or generic). Input schemas are either missing entirely or lack proper type definitions and parameter descriptions. No output schemas are documented. Error handling is absent. The server appears to be a collection of shell scripts and Python utilities without formal MCP tool registration or schema enforcement. Tools like 'browser_close', 'check_for_updates', and 'filesystem_watcher' have descriptions but no visible input validation or structured output documentation. The mcp_* tools (mcp_list, mcp_call, mcp_emit) are meta-tools for invoking other MCP servers, but their own schemas are incomplete. No evidence of idempotency, permission gating, or recovery guidance.
Close the browser context in Playwright MCP server
Check for new items to process from various sources (Gmail, WhatsApp, filesystem)
Create a markdown action file in the Needs_Action folder with frontmatter metadata and suggested actions
Monitor a drop folder for new files and create action items for processing
Get all pending action files in Needs_Action folder, sorted by priority (urgent first)
Invoke Qwen Code to process an action file with AI analysis and task completion
Call a tool on an MCP server with specified parameters
No input schemas visible for 6 of 10 tools (browser_close, check_for_updates, get_pending_files, invoke_qwen, and others). Tools with empty {} input or no schema definition cannot guide LLM parameter construction.
Descriptions are too short or generic. 'Close the browser context in Playwright MCP server' (8 words) and 'Check for new items to process from various sources' lack WHEN to use, prerequisites, and return value guidance. Baseline: 194 chars average for A+ tools.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields. E.g., get_pending_files returns 'all pending action files' but structure is unknown, does it return file paths, metadata objects, or raw content?
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | <=2025-11-25 | v2 |
Emit tool schemas as markdown or JSON from an MCP server
List available tools from an MCP server (HTTP or stdio transport)
Update Dashboard.md with current workflow statistics and recent activity
No error handling or recovery guidance. Tools fail silently or with generic errors. E.g., invoke_qwen has no documented failure modes, retry logic, or guidance for 'action file not found' scenarios.
Parameter descriptions missing or incomplete. 'item' in create_action_file is described as 'Item returned from check_for_updates()' but the structure of 'item' is not defined. 'vault_path' and 'drop_folder' in filesystem_watcher lack format/constraint guidance.