Cloud-hosted MCP server that wraps Microsoft's Markitdown library for converting documents to Markdown. Supports 29+ file formats including PDF, DOCX, PPTX, XLSX, and images. Ideal for RAG pipelines, knowledge bases, and AI workflows.
Single tool with moderate quality issues. Tool name lacks clear action verb (convert_to_markdown_tool is passive/descriptive rather than action-driven like 'convert_document_to_markdown'). Description is present and reasonably detailed (122 chars), meeting baseline expectations. Input schema is visible with proper types (string parameters), but both parameters are marked 'required: false' which creates ambiguity, the tool description says 'Either...must be provided' but the schema doesn't enforce this constraint. Parameter descriptions are present and adequate (29-41 chars each). No output schema documented. Error handling is implemented in the underlying function but not exposed in the tool definition itself. The tool performs a single, clear responsibility (document-to-markdown conversion) which is good composition, but lacks pagination support and result limits for potentially large markdown outputs.
Convert documents (PDF, DOCX, PPTX, XLSX, images, etc.) to clean Markdown. Supports 29+ file formats. Ideal for RAG pipelines, knowledge bases, and AI workflows.
Tool name lacks action verb. 'convert_to_markdown_tool' is descriptive but not action-first. Should be 'convert_document' or 'convert_file_to_markdown' to match pattern verb_noun naming and let LLMs infer intent from name alone.
Input schema marks both parameters as 'required: false' but tool description requires at least one. LLMs may pass neither, invoking an error. Schema should enforce the mutual-exclusivity constraint explicitly (e.g. via oneOf, not just description text).
No output schema documented. Tool returns raw markdown string, but LLMs need to know expected structure, length limits, and content format. Should document: 'Returns markdown text (string, typically 500 - 50000 chars depending on input document).'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 43 | - | v1 |
Parameter names are snake_case in schema (file_url, file_base64) but camelCase in the async function (fileUrl, fileBase64). Inconsistency may confuse LLMs or agents relying on schema vs actual parameter mapping.
No result pagination, limiting, or streaming for large documents. A 50MB PDF could produce 100k+ chars of markdown, consuming the agent's entire context window. Tool should cap output or support chunking.
Error handling is implemented in the async function (ValueError, HTTPStatusError, etc.) but not exposed to the tool schema. LLMs have no way to know what errors are possible or how to recover. Schema should include error documentation.