Loading ToolBench

Reading the current corpus…

mcp-clickhouse-logs MCP server · F · ToolBench

← Browse

mcp-clickhouse-logs

github.com/nfedotenko/mcp-clickhouse-logs

An MCP server (stdio transport) exposing read-only ClickHouse querying and cross-table log correlation tools.

2026-07-28+STDIO2 tools

47/100

Definition Quality53

Protocol Readiness50

Supportability35

Spec posture

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

Static source inference · medium confidence

Current-spec patterns detected

Summary

Two tools with clear, domain-specific descriptions (194 - 250 chars, within baseline 34 - 392). Both are read-only and well-scoped. However, input schemas are MISSING from the visible code, only parameter names and types are inferred from the description text. No explicit JSON Schema definitions are shown in internal/tools/tools.go. Parameter descriptions exist in the tool descriptions but not in a structured schema. Output schemas are undocumented. Error handling and recovery guidance are absent. Naming is verb-forward (clickhouse_query, correlate_logs) and appropriate for the domain.

Tools (2)

clickhouse_queryread onlysource verified52/100

Execute a read-only SQL query (SELECT, SHOW, DESCRIBE, EXPLAIN) against ClickHouse and return the rows as JSON. Queries run with readonly=1 and a server-side row limit.

correlate_logsread onlysource verified53/100

Fetch log entries for one session_id across the configured log tables within a time window around a pivot time, merged and sorted by timestamp. Use it to reconstruct what happened to a session across services.

Issues (5)

CRITICALtool ↗

Input schemas are not visible in source code. Tools are registered with mcp.Tool{Name, Description} only; no InputSchema field is populated. Per HARD SCORING RULE, schema score must be 0 for both tools.

clickhouse_querycorrelate_logs

HIGHresponse-shaper ↗

Output schemas are completely undocumented. LLMs cannot plan downstream calls or extract fields without knowing the structure of QueryResult or correlated log entries.

clickhouse_querycorrelate_logs

HIGHrecovery-guide ↗

No error handling or recovery guidance. If a query fails, times out, or returns no rows, the LLM has no actionable next step. Descriptions lack 'When to use' and 'What to do if' guidance.

clickhouse_querycorrelate_logs

MEDIUMconstrained-input ↗

clickhouse_query parameter 'query' lacks format/constraint documentation. No mention of allowed SQL keywords, injection safeguards, or what happens if readonly=1 is violated.

Recommendations

Add explicit InputSchema to both tools in mcp.Tool registration. Define 'query' as {type: 'string', description: '...'} and 'session_id', 'around', 'window', 'tables', 'limit' with proper types, enums (where applicable), and constraints.
Document output schemas in tool descriptions or as a separate schema field. E.g. 'Returns {columns: [string], rows: [{[key: string]: any}]}' for clickhouse_query and '{entries: [{timestamp, session_id, table, message, ...}], count: int}' for correlate_logs.
Add error handling guidance: 'If query times out, reduce the time window or add LIMIT. If no rows found, verify table names with SHOW TABLES. If readonly violation, contact admin.'
Specify parameter constraints in descriptions: 'query must be SELECT, WITH, SHOW, DESCRIBE, EXPLAIN, or EXISTS; no INSERT/UPDATE/DELETE. around must be RFC3339 format (e.g. 2026-08-09T12:00:00Z). window must be a Go duration (e.g. 5m, 1h, 30s).'
Add per-parameter validation examples: 'If tables is empty, all configured tables are searched. If limit exceeds max_rows config, it is capped server-side.'

Score history

Overall score trend

First recorded score · v2 rubric

47/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	47	2026-07-28+	v2