Three tools with basic schemas and descriptions, but significant gaps in parameter documentation and output specification. Tool names follow verb_noun convention (create_, add_, connect_) which is good. However, descriptions are minimal (10-50 chars), most parameters lack descriptions, and no output schemas are documented. The server uses in-memory state (diagrams Map) with no persistence or error recovery guidance. No tool annotations (destructiveHint, idempotentHint) despite all three being WRITE operations. Error handling is absent, no guidance on what happens if a diagram ID is invalid or element addition fails.

• Add output schema documentation for each tool. E.g., create_bpmn_diagram should return {diagramId: string, createdAt: ISO8601, xml: string}. This enables downstream tool chaining and LLM planning.
• Expand tool descriptions to 80-150 chars with actionable context. E.g., 'Create a new BPMN diagram with an empty process. Returns a diagram ID for use with add_bpmn_element and connect_bpmn_elements. Diagrams are stored in-memory and lost on server restart.'
• Add descriptions to all parameters. For elementType enum, explain the semantic difference between UserTask (manual work), ServiceTask (automated), and ScriptTask (code execution). For x/y coordinates, specify units (pixels? canvas units?) and valid ranges.
• Implement error handling with recovery guidance. E.g., if diagramId not found, return 'Diagram not found. Call create_bpmn_diagram() first to initialize a diagram.' If elementType is invalid, return 'Invalid elementType. Must be one of: bpmn:StartEvent, bpmn:EndEvent, ...'
• Add tool annotations: mark all three tools with destructiveHint=true and idempotentHint=false (since repeated calls may duplicate elements). This signals to LLMs that retries are risky and confirmation is needed.