Model Context Protocol server for Logchef - provides tools for querying logs, managing sources, profiles, and collections in a Logchef instance
Scoring was not performed
Output schemas not documented. Tools return results but no specification of response fields is visible. LLMs cannot plan downstream actions without knowing what fields to extract.
query_logs description lacks guidance on when to use this vs get_log_histogram, expected format of results, or what constraints the LLM must follow. ClickHouse SQL examples are shown but no warning about injection risks or statement restrictions.
No error recovery guidance. Tools do not document what errors are retryable, user-fixable, or fatal. If query_logs fails due to malformed SQL or timeout, the LLM has no guidance on recovery steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | F | 22 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 58 | - | v1 |
Tool annotations missing despite clear risk levels. delete_collection is DESTRUCTIVE but no destructiveHint visible. get_* tools are READ_ONLY but no readOnlyHint visible. This prevents MCP clients from applying appropriate safeguards.
query_logs parameter 'raw_sql' accepts freeform ClickHouse SQL with no validation constraints documented. LLMs could inject malicious statements. No mention of what statements are forbidden or allowed.
get_log_histogram description uses example values like '2025-06-12 05:00:00' in parameter descriptions. LLMs may reuse these literal dates in actual calls.
No pagination support documented for list_* tools. get_sources and get_collections have no limit, offset, or cursor parameters. If there are many sources/collections, the response could be massive and waste tokens.