MCP server for capturing and analyzing SSH session logs. Provides tools to list, search, and monitor SSH sessions with ANSI stripping and efficient polling.
Four tools with clear verb-noun naming (list_, focus_, read_, search_) and reasonable descriptions (100-200 chars). All tools have input schemas with typed parameters and descriptions. However, output schemas are partially documented (focus_session and read_session_since return dicts but lack formal schema declarations), and error handling is minimal, errors return bare dicts without recovery guidance. No tool annotations (readOnlyHint, idempotentHint) despite all being read-only. Parameter descriptions are adequate but lack format constraints (e.g., regex pattern for hostname, byte_offset range). Missing pagination support for list_sessions and search_session despite potentially large result sets.
Read the most recent session log for a hostname. Returns the last N lines of clean (ANSI-stripped) output plus a byte_offset you can pass to read_session_since for efficient polling. Hostname defaults to the server's --hostname value when not provided.
List captured SSH sessions. When the server is started with --hostname it lists only that host's sessions.
Return only new output since the last read (pass byte_offset from focus_session or a previous read_session_since call). Use this for polling / /loop watch mode. Hostname defaults to the server's --hostname value when not provided.
Search all session logs for a hostname using a Python regex pattern. Useful for finding errors, specific commands, or events across the full history. Hostname defaults to the server's --hostname value when not provided.
Output schemas not formally documented. Tools return dicts with fields like 'byte_offset', 'session_start', 'new_content', but no JSON Schema definition visible. LLMs cannot reliably extract fields or plan downstream calls without explicit output structure.
No tool annotations despite all tools being read-only. FastMCP supports readOnlyHint via @mcp.tool(). Adding this signals to agents that these tools are safe to call without side effects.
Error handling returns bare dicts with 'error' key but no recovery guidance. E.g., 'hostname is required' does not tell the LLM to call list_sessions() first. Errors should include actionable next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Parameter 'hostname' lacks format constraints. Description says 'defaults to server's --hostname value' but does not specify valid format (e.g., FQDN, alphanumeric + hyphens). LLMs may pass invalid hostnames.
search_session returns up to 50 matches by default but no pagination support. If a hostname has thousands of matching lines, the result is truncated silently. No next_cursor or offset mechanism to retrieve additional matches.