Loading ToolBench
MCP server for GOD CRM - Universal Workspace Management System. Provides tools for AI agents to interact with CRM data, execute code, manage approvals, and access BDD specifications.
Critical gaps across naming, descriptions, and schemas. Tool names lack consistent verb prefixes (run_code_loop, manage_plan are vague). Descriptions range from adequate (run_code: 180 chars) to severely underdeveloped (get_workspace_info: 50 chars, well below 100-char minimum). Most parameters lack type constraints and detailed descriptions. Destructive tools (delete_row, execute_sql, write_file) have no error handling guidance or confirmation patterns. No output schemas documented. Security concerns: execute_sql and run_migration expose dangerous operations without permission gates or audit trails. Composition issues: run_code and run_code_loop duplicate functionality; manage_plan mixes task creation and updates in one tool.
Analyze table data and generate insights
Create a new dashboard in the workspace
Delete a single row from a table
Delete multiple rows from a table
Execute SQL queries against the database
Get the schema of a table including column definitions
Get workspace information including projects, tables, and configuration
List BDD specifications and acceptance criteria for a document
Destructive tools (delete_row, delete_rows, execute_sql, write_file, run_migration) lack confirmation patterns, dry-run support, and error recovery guidance. No audit trail or permission gates documented.
Descriptions severely underdeveloped. 10 of 16 tools have descriptions under 100 characters (baseline 100 - 1024). Examples: get_workspace_info (50 chars), execute_sql (20 chars), delete_row (25 chars). LLMs cannot determine when to select these tools.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | <=2025-11-25 | v2 |
List all tables in a project or space
Create or update a plan for the current conversation. Use this when a task requires 3+ steps. Maximum 20 tasks per plan.
Query data from a table with filtering and pagination
Execute code in various programming languages. Supported languages: python, javascript, typescript, cpp, java, go, rust, ruby, php, csharp, bash, sql, kotlin, swift, r, perl, lua. Returns stdout, stderr, execution time and memory usage. Use for quick calculations, data processing, or testing code snippets.
Execute code and automatically continue the conversation with the result. Use for iterative tasks: debugging, experiments, step-by-step calculations. The result will be analyzed automatically and the conversation will continue.
Run database migrations
Validate code syntax without executing it. Fast check for syntax errors before running. Works best with Python (uses AST parser).
Write content to a file on the filesystem
Parameter descriptions missing or trivial. Most parameters lack actionable guidance on format, range, or constraints. Example: 'analysis_type' in analyze_table_data has no enum or description of valid values.
Duplicate/overlapping tool functionality. run_code and run_code_loop both execute code; manage_plan combines create, update, add, remove in one tool. Violates single-responsibility principle.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields. Example: query_table_data returns data but no schema for result structure, pagination, or field types.