Any file → Markdown, for MCPG. One cdylib carrying two entities over a single conversion engine: a `backend` binding exposing `convert_to_markdown` as an MCP tool, and a `transform` entity that converts a document already present in a payload as a global transform or a pipeline `plugin_transform` step. Reads Office files, PDF, HTML, e-mail, archives, images and audio and emits LLM-friendly Markdown, with optional operator MiniJinja templates and optional image captioning / transcription dispatched to an existing LLM binding.
Single tool 'convert_to_markdown' has a well-structured input schema with 6 parameters, all typed and described. Tool description is comprehensive (180+ chars) and explains the conversion capability, input formats, and enrichment dispatch. However, output schema is not documented in the visible code, the tool returns converted Markdown but the response structure is not specified. Parameter descriptions are clear but lack format constraints (e.g., valid MIME types, charset values). No error handling guidance visible. The tool follows verb_noun naming and accepts multiple input formats (base64, text, URI), which is good for usability. Missing: output schema documentation, error recovery guidance, and parameter value constraints (enums for mimetype/charset).
Convert a document to Markdown. Accepts content as base64-encoded bytes, plain text, or a URI (mcpg-resource://, data:, or https:// when enabled). Returns LLM-friendly Markdown with optional enrichment (image captions, OCR, audio transcription) dispatched to an LLM binding.
Output schema not documented. Tool description does not specify the structure of returned Markdown (fields, types, nested objects). LLMs cannot plan downstream operations or extract specific fields without knowing the response shape.
Parameter 'mimetype' and 'charset' lack enum or format constraints. Descriptions say 'Declared content type' and 'Character set' but do not list valid values (e.g., 'utf-8', 'iso-8859-1') or reference standards. LLMs may pass invalid values.
No error handling guidance visible in tool definition. If conversion fails (unsupported format, corrupted file, network timeout on URI fetch), the tool description does not explain what error the LLM should expect or how to recover.
Transport mechanism unknown. Cargo.toml and source code do not reveal whether this plugin uses HTTP, SSE, STDIO, or a custom mcpg-plugin-protocol transport. Cannot assess remote accessibility or protocol compliance without this.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Tool accepts URIs (mcpg-resource://, data:, https://) but no description explains rate limits, timeout behavior, or maximum file size. Large file fetches could hang the agent or exhaust memory.