Loading ToolBench
An MCP agent that transforms high-level business intents into structured Data Product Requirement Prompts through conversational refinement
The server defines 3 tools with explicit schemas, descriptions, and parameter validation. However, there are significant gaps in parameter descriptions, output schema documentation, and error recovery guidance that prevent a higher score. Naming is adequate but not exemplary. Tool definitions are explicitly visible in src/data_planning_agent/mcp/tools.py, so all tools can be scored on actual evidence rather than inference.
Continue an existing planning conversation by providing responses to questions. The agent will ask follow-up questions or indicate when requirements are complete.
Generate a complete Data Product Requirement Prompt (Data PRP) from a session. Creates a structured markdown document with executive summary, business context, data requirements, and success criteria. Optionally saves to a file.
Start a new planning session to gather requirements for a data product. Provide an initial business intent, and the agent will ask clarifying questions to refine the requirements. Returns a session ID and initial questions.
No output schemas documented for any tool. LLMs cannot predict what fields are returned, preventing composition of multi-step workflows and forcing agents to make assumptions.
Parameter descriptions lack format constraints, ranges, and validation rules. E.g., session_id has no documentation of valid format; output_path has no guidance on required directory structure or filename; user_response has no guidance on max length.
Tool descriptions are missing critical context: dependencies between tools (start_planning_session must precede continue_conversation), expected behavior on errors (invalid session_id), and when each tool should be used.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 46 | - | v1 |
No error handling guidance in tool descriptions. If session_id is invalid, should the agent retry? Ask the user? What error messages can be expected?
generate_data_prp is marked WRITE risk but lacks idempotency documentation. If called twice with the same session_id, does it overwrite the file? Append? Error? This ambiguity invites duplicate outputs or failures.
Acronym 'Data PRP' in descriptions and tool name may be unclear to LLMs without context. Consider expanding to 'Data Product Requirement Prompt' or providing the full form in the description.