Loading ToolBench
MCP server for orchestrating CRM relationship graph analysis and deal management through agentic workflows
This server exhibits critical gaps across all definition quality dimensions. Tool definitions are incomplete, descriptions are generic, and schemas are not explicitly documented. The source code shows only minimal tool registration without proper MCP-compliant schema definitions. Tool 'execute_skill_action' is mentioned in mcp_server.py but has no visible input schema, description, or parameter documentation. Tool 'orchestrate_deal_graph' has a basic input schema in client.py but lacks formal JSON Schema structure with type definitions. No output schemas are documented for either tool. Error handling is absent, there are no recovery guides, categorization, or actionable error messages. No security patterns are implemented (no permission gates, audit trails, or input validation visible). The example_usage.py shows direct Python imports rather than MCP protocol invocation, suggesting this is not a proper MCP server implementation at all.
Execute a skill action in the agentic CRM relationship graph orchestrator
Orchestrate the deal graph by analyzing contact emails and deal stage to determine relationship strength, next best step, and synced deal value
Tool 'execute_skill_action' has no visible input schema, parameters, or description. Source shows only a tool name in a JSON config with no formal definition.
No input schemas use JSON Schema format with proper type definitions. The 'orchestrate_deal_graph' input is described informally in Python docstring-like comments, not as a structured schema MCP clients can parse.
No output schemas are documented. Tools return Python dicts (e.g. 'relationship_strength_score', 'next_best_step', 'synced_deal_value_usd') but there is no formal schema declaring these fields, types, or whether additional fields may be present.
Description for 'execute_skill_action' is generic ('Execute a skill action in the agentic CRM relationship graph orchestrator') and does not explain WHEN to use it, what ACTIONS it accepts, or what it RETURNS. LLMs cannot determine when to select this tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 22 | 2026-07-28+ | v2 |
No error handling patterns are visible. If a tool call fails (invalid email, missing deal data, API timeout), there are no recovery guides, error categorization, or actionable error messages to guide agent retry logic.
Parameter descriptions are minimal. 'contact_emails' is described as 'List of contact email addresses to include in the relationship graph analysis' but does not specify format (single email, array of emails), whether emails must be validated, or what happens if an email is invalid.
No input validation constraints are documented. 'deal_stage' accepts a string with a default 'DISCOVERY', but there is no enum of valid deal stages, no regex pattern, and no description of allowed values. LLMs will hallucinate invalid stages.
No security patterns implemented. Tool definitions do not declare permissions, there is no audit trail visible, and no input sanitization is evident. A tool analyzing contact emails and deal values should gate access and log who called it.
Tool 'execute_skill_action' naming is ambiguous. Does it execute a single action, multiple actions, or a workflow? Does it create, update, delete, or query? The name does not follow verb_noun convention and does not convey intent.