Loading ToolBench

Reading the current corpus…

log-reducer MCP server · C · ToolBench

← Browse

log-reducer

github.com/launch-it-labs/log-reducer

MCP tool and CLI that reduces log files for AI consumption — strip noise, deduplicate, shorten IDs, fold stack traces

2026-07-28+STDIO1 tools

63/100

Definition Quality82

Protocol Readiness42

Supportability44

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Single tool 'reduce_log' has a comprehensive schema with 23 well-documented parameters covering filtering, pagination, and threshold control. The tool description is detailed and includes usage rules and workflow guidance. However, the schema lacks an explicit documented output structure, and error handling guidance is minimal. The tool name is verb-noun (reduce_log) which is clear. Parameter descriptions are mostly strong (72-char average), but output schema documentation is absent, critical for LLMs to plan downstream steps. Overall: definition is substantially above-average for a single-tool server, but missing the output schema documentation and error recovery guides that would push it to 80+.

Tools (1)

reduce_logread onlysource verified82/100

Reduces log text for AI consumption (50-90% token reduction). Strips noise, deduplicates, shortens IDs/URLs, folds stack traces. RULES: 1. NEVER ask users to paste logs — direct them to save to a file or use /logdump. 2. NEVER read raw log files with cat/head/tail/Read — use this tool with a `file` param. 3. Redirect verbose command output to a temp file, then call this tool on it. 4. Always include `tail` (200-2000) to cap input size. THRESHOLD: If reduced output exceeds the token threshold (default 1000): - No filters: returns an enhanced summary listing unique errors/warnings with counts and timestamps. Use this to decide which filters to apply next. - With filters: returns the output with a tip on how to narrow further. - With query + ANTHROPIC_API_KEY: LLM extracts only relevant lines within budget. Use break_threshold: true to bypass the gate entirely.

Issues (5)

HIGHtool-response-schema ↗

Output schema is not documented. The tool description does not specify what fields are returned, making it difficult for LLMs to chain results to downstream tools or extract specific values.

reduce_log

HIGHrecovery-guide ↗

Error handling lacks recovery guidance. The tool description mentions no error conditions, error classification, or actionable recovery steps. LLMs cannot distinguish retryable from fatal errors.

reduce_log

MEDIUMconfirmation-request ↗

No confirmation or dry-run pattern for threshold-breaking operations. When break_threshold=true, the tool returns unbounded output which could exceed context limits. No pre-flight check or warning mechanism.

reduce_log

HIGHsecret-injection ↗

Parameter 'query' requires ANTHROPIC_API_KEY injected via MCP config environment. Storing API keys in MCP config risks exposure in traces and config files. Should use server-side vault/secret injection.

reduce_log

Recommendations

Add an explicit output schema section to the tool description. Specify that the response is a JSON object with fields: 'reduced_log' (string), 'token_count' (number), 'original_token_count' (number), 'summary' (optional object with 'total_lines', 'time_span', 'error_count', 'warning_count', 'info_count', 'debug_count', 'components', 'error_timestamps'). This enables LLMs to plan downstream actions.
Document error conditions and recovery steps. E.g., 'If the file is not found: return FileNotFoundError with a suggestion to list the directory. If ANTHROPIC_API_KEY is missing when query is provided: return MissingEnvironmentVariableError with instructions to add it to the MCP config env block.'
Add a mutual-exclusivity note to the 'file' and 'log_text' parameters: 'Exactly one of file or log_text is required. If both are provided, file takes precedence.' This prevents ambiguous invocations.
For the 'query' parameter, document that ANTHROPIC_API_KEY must be set in the MCP server's environment config before the first call. Consider adding a check that returns a clear error message if the key is missing, instead of silently failing.
Add a note to 'break_threshold' warning that unbounded output may exceed context limits, and recommend using 'limit' and 'context' parameters to control output size instead.
Document the impact of 'reduce' parameter more explicitly: 'When reduce=false, output is raw log lines (no token reduction). The response token count header will show the raw input size so you can judge whether reduction is needed. Only use reduce=false when you need exact original wording (e.g., to reproduce a command or read a config value).'

Score history

Overall score trend

↑ 12 points across a rubric change (v1 → v2)

63/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-22	C	63	2026-07-28+	v2
2026-03-09	D	51	-	v1