MCP server for managing and querying Tiger Cloud database services (managed TimescaleDB/PostgreSQL). Provides tools for provisioning, forking, starting/stopping/resizing/deleting instances, rotating credentials, fetching logs, executing SQL queries, and searching documentation.
Tiger MCP has 17 tools with complete names, descriptions, and input schemas visible in internal/mcp/server.go and related files. Naming follows verb_noun convention well (service_list, service_get, service_create, etc.), though there is tool-name repetition risk (14 of 17 start with 'service_'). Descriptions are present for all tools and range from 60-150 characters, meeting minimum length requirements. Input schemas are properly typed with JSON Schema format. However, OUTPUT schemas are not documented anywhere in the provided code, tools return results but the response structure is not specified, forcing LLMs to infer output fields. Error handling is mentioned as a feature but specific recovery guidance is not visible in tool definitions. Parameter descriptions are adequate but lack constraints (enums, format strings, min/max bounds) for many fields. The db_query tool exposes a 'query' parameter marked 'sensitive' but does not restrict it server-side, relying on client-side handling. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are visible despite the risk classification being present (READ_ONLY, WRITE, REVERSIBLE, DESTRUCTIVE).
Execute a SQL query against a Tiger Cloud database service
Get the schema of a Tiger Cloud database
Submit feedback about the Tiger MCP server
List backups for a Tiger Cloud database service (experimental/preview feature)
Create a new Tiger Cloud database service
Delete a Tiger Cloud database service
Fork an existing Tiger Cloud database service
No output/response schemas documented for any tool. LLMs cannot infer what fields the response contains, forcing them to guess at downstream data dependencies and chain calls incorrectly.
14 of 17 tools start with 'service_', creating naming ambiguity. LLMs struggle to disambiguate similar-prefixed tools. No clear distinction between service_metrics_available vs service_metrics_series vs service_backups in naming alone.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 67 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 45 | - | v1 |
Get details about a specific Tiger Cloud database service
List all Tiger Cloud database services in the current project
Fetch logs from a Tiger Cloud database service
Get available metrics for a Tiger Cloud database service (experimental/preview feature)
Fetch metrics time series data for a Tiger Cloud database service (experimental/preview feature)
Rename a Tiger Cloud database service
Resize a Tiger Cloud database service to a different compute tier
Start a stopped Tiger Cloud database service
Stop a running Tiger Cloud database service
Update the password for a Tiger Cloud database service
Input parameters lack constraint documentation. 'region_code' and 'service_type' in service_create have no enum values listed; 'new_compute_unit' in service_resize has no valid options; 'lines' in service_logs has no min/max bounds. LLMs will hallucinate invalid values.
db_query tool accepts 'query' parameter marked 'sensitive' but no input sanitization, parameterization, or injection prevention hints are documented. SQL injection risk if LLM constructs queries with untrusted input.
Tool risk classification (READ_ONLY, WRITE, REVERSIBLE, DESTRUCTIVE) is recorded but NOT exposed as tool annotations (readOnlyHint, destructiveHint, idempotentHint in MCP). LLMs cannot see safety hints and may misplan destructive operations.
Destructive operations (service_delete) have no confirmation/dry-run mechanism documented. An agent could delete a production service without a reversibility step or warning.
No pagination support visible in service_list or other list-returning tools. Descriptions do not mention limits, cursors, or offset/limit parameters. Large result sets will blow context windows.
Error handling is mentioned as a feature (errorReporting=true) but specific recovery guidance is not visible in tool descriptions. Tools do not include actionable error messages like 'User not found. Try search_users() first.'