Loading ToolBench
MCP server for creating, reading, and manipulating Microsoft Word documents with support for Azure Blob Storage
Server provides 24 tools with schemas and descriptions present, but quality is inconsistent. Naming is generally verb-first and clear (add_heading, get_document_text, create_document), which is good. However, descriptions lack depth and LLM-optimization guidance. Parameter schemas are present with type definitions, but many descriptions are minimal (e.g., 'Heading text' for the text parameter in add_heading). Output schemas are not documented anywhere in the visible code. Error handling returns plain strings rather than structured errors with recovery guidance. Security considerations are weak, no permission checks visible, no audit logging, and file path handling via storage functions is abstracted but not clearly validated. Tool composition is reasonable (separate tools for different operations), but some tools like format_text are overly parameterized (11 parameters) without clear guidance on which are mutually exclusive or dependent. The server handles Azure Blob Storage integration, which is good for production use, but this is not reflected in parameter documentation or security declarations.
Add a document as a template to the template library. Automatically cleans the document by: keeping title row (row 0) and header row (row 1) in all tables, removing all data rows; extracting and returning table titles for user reference; all other content (text, formatting, customer success info, etc.) is preserved as-is.
Add a heading to a Word document.
Add a paragraph to a Word document.
Add an image to a Word document.
Add a table to a Word document.
Add a row to an existing table in a Word document.
Output schemas not documented. No tool response format is defined in descriptions or visible schema declarations. LLMs cannot plan downstream operations or know what fields to extract from results.
Error handling returns plain text strings instead of structured error objects with categorization and recovery guidance. Errors like 'Document does not exist' give LLMs no guidance on whether to retry, look up the name, or escalate.
format_text has 11 parameters without documenting which are mutually exclusive or dependent. LLMs will guess at combinations (e.g., passing bold=true AND italic=true when only one is valid). Parameter descriptions should state 'At least one formatting option required' or clarify mutual exclusivity.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Clean up expired documents from storage.
Create a copy of a Word document.
Create a custom style in the document.
Create a new Word document with optional metadata.
Create a new document from a template with variable substitution.
Delete a template from the template library.
Get download URL for a document.
Format a table with borders, shading, and structure.
Format a specific range of text within a paragraph.
Get information about a Word document.
Get the structure of a Word document.
Extract all text from a Word document.
Get the raw XML structure of a Word document.
Get detailed information about a specific template.
List all .docx files in storage.
List all available document templates.
Merge multiple Word documents into a single document.
Apply shading/filling to a specific table cell.
No tool annotations present (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot determine which tools are safe to retry (idempotent), which modify data (destructive), or which only read (readOnly). This is critical for agent planning and safety.
Permission and security declarations missing. No scope declarations (e.g., 'read:document', 'write:document', 'admin:template'). No visibility into who can call which tool or what audit trails are maintained. cleanup_expired_documents and delete_document_template are destructive but lack permission gates.
Parameter descriptions are minimal. Many say only 'Path to the Word document' or 'Heading text' without explaining constraints, valid ranges, or expected formats. Per the rubric baseline, average param annotation length in A+ tools is 72 chars; most here are <40 chars.
cleanup_expired_documents takes no parameters but is destructive. No confirmation step, dry-run option, or undo mechanism. An agent could accidentally wipe all old documents.
Tool naming: get_document_xml_tool is awkwardly named ('_tool' suffix is redundant). Prefer get_document_xml or extract_document_xml. Current name creates confusion about what the tool is.