Simple MCP server for testing Warden sandbox with basic filesystem operations
Four tools with basic schemas and minimal descriptions. All tools have input schemas with required parameters and type definitions, but descriptions are generic (10-30 chars) and lack context for LLM selection. No output schemas documented. Parameters lack format constraints, ranges, or validation guidance. No error handling patterns visible. Tool names follow verb_noun convention but descriptions do not explain WHEN to use each tool or what distinguishes them. Missing idempotency, pagination, and composition guidance.
List files in a directory
Read contents of a file
Test network connectivity
Write contents to a file
Descriptions are too short (10-30 chars) and lack context. 'Read contents of a file' does not explain when to use read_file vs list_directory, what formats are supported, or error recovery. LLMs cannot distinguish tools with minimal descriptions.
No output schemas documented. Callers do not know what fields read_file returns (string content? object with metadata?), what list_directory returns (array of strings? objects with size/type?), or what test_network returns (boolean? latency?). This forces LLMs to guess and breaks downstream tool chaining.
write_file lacks idempotency guidance and confirmation pattern. Writing to a file is destructive, the description does not warn that repeated calls overwrite data, nor does it offer a dry-run or confirmation step. Agents may accidentally overwrite critical files.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
path parameter lacks format constraints. No guidance on absolute vs relative paths, path traversal prevention, or allowed directories. LLMs may pass '../../../etc/passwd' or other malicious paths without validation hints.
No error handling guidance. If read_file fails (permission denied, file not found, encoding error), the tool does not tell the LLM what to do next. No recovery patterns, retryability classification, or actionable error messages.