MCP文件内容读取服务 - 从各类文件中提取内容。支持PDF、Office文档、OpenDocument格式、文本文件、压缩文件等多种格式的本地文件读取和转换
This MCP server has 2 tools with schemas visible in the code. Both tools have descriptions in Chinese and parameter schemas defined. However, critical issues emerge: (1) Both tool descriptions are under 40 characters when translated/considered in detail, the rubric requires 10-1024 chars with context about WHAT, WHEN, and WHY. (2) Parameter descriptions lack actionable constraint details, 'file_path' just says 'absolute path' without explaining validation, allowed directories, or recovery paths. (3) No output schemas are documented, the code shows TextContent responses but LLMs cannot see what fields are returned. (4) Error handling is absent from visible code, no guidance on retryability or recovery. (5) Tool names are acceptable ('read_local_file', 'convert_local_file') but descriptions do not explain the difference clearly or guide LLM selection. (6) The descriptions are in Chinese, which may limit LLM comprehension in English-dominant contexts. The server implements basic functionality but lacks the production-grade polish required for confident agent deployment.
将本地文件转换为markdown文件并保存到文件系统
读取本地文件系统中的PDF、Office文档(不支持图片)
Output schemas are not documented. Neither tool documents what the LLM receives in response. 'read_local_file' appears to return TextContent but structure is not explicit. 'convert_local_file' should return a filepath or status but this is invisible to the agent. LLMs cannot plan downstream calls or extract needed fields without explicit output schemas.
Parameter descriptions lack actionable constraints and validation details. 'file_path' does not explain: what directories are allowed? What happens if a path outside allowed_directories is passed? Can the agent recover? 'max_size' lacks context: is 20 MB a hard limit or advisory? What error occurs if exceeded? The rubric requires descriptions to state expected format, range, and allowed values.
Tool descriptions are in Chinese, which may limit comprehension by English-language LLMs and reduces portability across non-Chinese-speaking teams. While not a correctness issue, it creates a barrier for adoption in multi-lingual environments.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 37 | - | v1 |
No error handling guidance. Code shows ReadResponse with FailureType enum but tool descriptions do not document failure modes or recovery paths. If a file is not found, cannot be read (permission denied), or exceeds size limit, what does the LLM see? How should it recover? The rubric requires error responses to guide the LLM on retryability and next steps.
Tool descriptions do not clearly distinguish use cases. Both 'read_local_file' and 'convert_local_file' operate on the same file types and accept the same inputs. A user saying 'I need to read this PDF' could use either tool. The descriptions must explicitly state: read_local_file returns the content inline; convert_local_file writes a markdown file to disk. Without this clarity, the LLM may select the wrong tool.
Destructive operation 'convert_local_file' with overwrite=true lacks confirmation mechanism. If overwrite defaults to false, that is safe. However, when an agent explicitly sets overwrite=true, there is no confirmation step to prevent accidental overwrites. The rubric recommends confirmation patterns for irreversible operations.
Risk annotations visible (READ_ONLY, WRITE) suggest the server implements toolAnnotations, but descriptions do not explain the consequences. A WRITE operation like convert_local_file must state clearly: 'This tool creates or overwrites files on disk. Use with caution.' Otherwise, agents do not understand the irreversibility.