MCP server for DBTrail — provides read-only tools for querying binlog events, recovering data, checking index status, and listing schema changes from a DBTrail index
bintrail-mcp defines 6 tools with complete input schemas and descriptions. Tool names follow verb_noun convention (query, recover, reconstruct, recover_cascade, status, list_schema_changes). All parameters have type definitions and descriptions. However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Parameter descriptions lack actionable constraints (e.g., RFC3339 format stated but no validation guidance). Error handling and recovery paths are not specified. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles (READ_ONLY vs REVERSIBLE). Descriptions are adequate (100-200 chars) but could better explain WHEN to use each tool vs alternatives.
List all schema changes (DDL events) recorded in the index, including ALTER TABLE, CREATE TABLE, DROP TABLE, and other DDL operations.
Query binlog events from the DBTrail index with optional filtering by schema, table, event type, time range, and GTID. Returns matching events with their metadata.
Time-travel query: reconstruct table state at a specific point in time using baseline snapshots and binlog events. Opt-in tool requiring baseline configuration.
Generate SQL reversal scripts to undo data changes. Takes binlog events and produces DELETE/INSERT/UPDATE statements to restore prior state.
Recover child table rows when parent rows are deleted. Synthesizes missing child events using indexed parent-child relationships and optional baseline snapshots.
Check the health and status of the DBTrail index, including schema version, partition information, and index statistics.
Output schemas not documented. LLMs cannot predict response structure, breaking tool chaining and forcing agents to guess field names for downstream calls.
Tool annotations missing. recover and recover_cascade are REVERSIBLE (destructive) but lack destructiveHint annotation. query, reconstruct, status, list_schema_changes are READ_ONLY but lack readOnlyHint. Agents cannot distinguish safe from risky tools.
Parameter constraints not actionable. RFC3339 format mentioned in descriptions but no validation guidance. index_dsn, baseline_dir, baseline_s3 accept free-form strings with no format validation hints. Agents cannot self-correct invalid input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
No error recovery guidance. Tools lack descriptions of failure modes (e.g., 'index not found', 'invalid GTID', 'baseline missing'). Agents cannot plan recovery steps.
Tool differentiation unclear. recover vs recover_cascade both generate reversal scripts but differ in scope (single table vs cascade). Descriptions do not explain when to choose one over the other.