Loading ToolBench
MCP server for querying browser console logs captured by the Console Log Copier Chrome extension
Server provides 11 tools with basic schemas and descriptions. Strengths: all tools have non-empty descriptions (>20 chars) and input schemas with type definitions. Weaknesses: descriptions are functional but generic (avg ~100 chars, well below 194-char baseline); no output schemas documented; no error handling guidance; parameters lack depth (no enums, no constraints, no format specifications); destructive tools (clear_session_logs, clear_url_logs, clear_all_logs) have no confirmation patterns or permission gates; tool composition issues (search_logs combines query+levels+categories with overlapping semantics); no security measures visible for database access control or audit trails. The codebase shows solid database structure and prepared statements, but the tool interface lacks polish and production-grade safeguards.
Delete all console logs and sessions from the database
Delete all logs for a specific session
Delete all logs for sessions matching a URL pattern
Export logs for a session in JSON or plain text format
List all active sessions with their tab IDs, URLs, titles, start times, and log counts
Retrieve all error-level logs from all active sessions, optionally filtered by URL
Get the most recent session (active or inactive) for a given URL pattern
Get information about the MCP server and database statistics
No output schemas documented. Tools return structured data from the database (rows, log objects, counts), but LLMs have no visibility into response field names, types, or structure. This forces LLMs to guess what fields exist and risks context loss between tool calls.
Destructive tools (clear_session_logs, clear_url_logs, clear_all_logs) lack confirmation/dry-run patterns and permission gates. An agent can delete all logs in the database with a single unconfirmed call. No audit trail or recovery mechanism visible.
No error handling guidance. Tools return database results or errors, but error responses lack recovery hints. If a session_id is invalid, does the tool return a 404, an empty result, or a descriptive error? Unclear. No guidance for LLMs on what to do next.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 23 | - | v1 |
Retrieve session history for a specific URL with optional limit on number of sessions returned
Retrieve all console logs for a specific session by session ID
Search for logs across all sessions matching a query string with optional level and category filters
Parameters lack constraints and enums. 'levels' in search_logs is a free-form array of strings, no enum of valid log levels provided (error, warn, info, debug, log, etc.). LLMs can hallucinate invalid values. 'format' in export_logs correctly uses an enum, but others do not.
Descriptions are generic and lack LLM guidance. Examples: 'Retrieve all console logs for a specific session by session ID' (55 chars), does not explain when to use this vs get_session_history, what data is returned, or whether results are paginated. No discovery hints or prerequisites stated.
No pagination support or result limits documented. search_logs defaults to limit=50, but no indication of maximum or whether pagination is supported. For a log aggregation tool expecting large result sets, lack of clear pagination guidance invites context-window exhaustion.
No security measures visible: no permission checks, no audit logging, no secret injection patterns. Any agent with access to the MCP server can clear all logs without authentication. No indication of who called what, when, or why.