DocumentMCP has critical definition gaps. Tool schemas are present but descriptions are minimal (10-50 chars, well below the 194-char baseline). The 'format' tool is registered as a prompt, not a tool, creating confusion. The 'summarize' tool appears twice with identical definitions but different file locations, suggesting incomplete refactoring. Parameter descriptions exist but are sparse. No output schemas documented. Error handling is minimal, tools raise ValueError without recovery guidance. The server lacks tool annotations (readOnlyHint, destructiveHint) despite having both read and write operations.

• Convert 'format' from @mcp.prompt() to @mcp.tool() so it can be discovered and chained by agents. Prompts are for multi-turn interactions, not tool composition.
• Consolidate duplicate 'summarize' definitions. Keep only one implementation; remove the duplicate from sampling/server.py or clarify the intended architecture.
• Expand tool descriptions to 100 - 200 characters. For edit_document: 'Modify a document by replacing exact text. Use this to make targeted edits; for full rewrites, use format. Returns success or error if text not found.'
• Document output schemas for all tools. E.g., read_doc_contents returns {type: 'string', description: 'The full document content'}; edit_document returns {type: 'object', properties: {success: boolean, message: string}}.
• Add recovery guidance to error messages. E.g., 'Doc with id {doc_id} not found. Available docs: deposition.md, report.pdf, financials.docx. Try read_doc_contents with one of these IDs.'

Inferred effective spec: <=2025-11-25.

• Relies on Sampling (deprecated) - integrate directly with the LLM provider API