An MCP server for cutting local and remote log files down to specified date ranges
Log Harvester defines 2 tools with partial schema visibility and uneven description quality. Tool names follow verb_noun convention ('cut_local_logs', 'cut_remote_logs') which is good. Both have descriptions, but they lack critical parameter details and error handling guidance. Input schemas are present and typed in the JSON decorators, but output schemas are not formally documented. The server handles warnings via a custom log handler but does not provide structured error recovery paths or actionable guidance for the LLM.
Cut local log files or directories down to the given date range.
Cut log files from a remote host (over SSH) down to the given date range.
Output schema not formally documented. Both tools return {'status', 'dest_path', 'warnings'} but return type is not declared in tool registration. LLM cannot predict structure or downstream tool compatibility.
Date format parameters (from_date, to_date) have examples in descriptions ('2025-10-09 15:30:00') but no formal format constraint (ISO 8601 regex or enum). LLMs frequently misformat dates; formal constraints are needed.
Error handling lacks recovery guidance. When warnings occur, the tool returns 'completed_with_warnings' status with a list, but does not explain what failed, why, or what the agent should do next (retry specific paths, investigate SSH timeout, check permissions, etc.).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
cut_remote_logs accepts hostname and port as optional overrides but does not validate them before passing to RemoteLogCutter. Invalid hostnames or ports could trigger SSH timeouts without clear error messages to guide the LLM.
SSH credentials are injected from Config().configs rather than passed as parameters, which is correct for security. However, there is no tool-level documentation of this security model, and no way for the LLM to verify or update SSH settings dynamically. The tool description should state 'SSH credentials are loaded from server config; use the TUI to update.'
Parameter 'log_paths' is described as accepting 'Local file or directory paths' for cut_local_logs and 'Remote file or directory paths' for cut_remote_logs, but does not specify: (1) Are symlinks followed? (2) What is 'non-recursively', does it scan subdirs or not? (3) What file extensions are supported? (4) What happens if a path doesn't exist?
Both tools modify state (write files to dest_path) but descriptions do not explicitly state this is a side-effecting, non-idempotent operation. Description should include 'This tool is NOT idempotent, multiple calls may overwrite dest_path files.' to inform agent retry strategies.