Loading ToolBench
MCP server for the ThoughtSpot REST API v2.0 (self-hosted, stdio)
Strong tool definitions with consistent naming (verb_noun pattern), comprehensive parameter descriptions, and clear risk classifications. All 14 tools have detailed descriptions (avg ~180 chars) and input schemas with typed parameters. However, output schemas are not explicitly documented in the source, and error handling guidance is minimal. Response format parameters are present but add noise. Tool composition is sound, each tool has a single responsibility and clear chaining via identifiers (connection_id, metadata_identifier, etc.).
Create a new data-warehouse connection (Snowflake, BigQuery, Redshift, Databricks, etc.). Supports optional validation and idempotent creation (if_not_exists).
Delete a data-warehouse connection by GUID or name.
Delete one or more metadata objects (Liveboards, Answers, Models, Connections, Logical Columns, or Logical Relationships) by GUID or name.
Export an Answer as a CSV, PDF, XLSX, or PNG file. Accepts either a saved Answer's metadata_identifier, or an unsaved Answer session's session_identifier + generation_number pair. The API returns raw file bytes, which this tool writes to output_path on the MCP server's local filesystem.
Export a Liveboard (and its visualizations) as a PDF, PNG, CSV, or XLSX file. Requires at least view access to the Liveboard, plus the relevant download privilege if RBAC is enabled. The API returns raw file bytes, which this tool writes to output_path on the MCP server's local filesystem.
Output schemas not documented. Tools return data but LLMs cannot see expected response structure (fields, types, pagination). Forces LLMs to guess at downstream field names and risks broken tool chains.
response_format parameter (markdown|json) present on most tools but adds noise and decision burden to LLM. Should default to json internally; if markdown needed, offer a separate export tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Fetch data from a saved Answer. Optionally apply runtime filters, sorts, or parameter overrides.
Retrieve a single connection by GUID or exact name, including full configuration and registered warehouse objects.
Fetch data from a Liveboard's visualizations. Optionally filter to specific visualizations or apply runtime filters/sorts.
Check whether the ThoughtSpot instance is reachable and credentials are valid. Performs a token exchange (unless a static token is configured) and a lightweight GET /system request.
Search for data-warehouse connections (Snowflake, BigQuery, Redshift, …) by identifier, name pattern, or warehouse type. Supports pagination and optional full details.
Run a TML search query against a Worksheet, View, Table, or SQL view. Requires at least view access to the data source; if RBAC is enabled, also requires the "Can download detailed data" privilege.
Search for metadata objects (Liveboards, Answers, Models, Connections, Tags, Users, ...). Wraps POST /metadata/search with a multi-strategy fallback scoped to the MODEL alias.
Enable (ACTIVATED) or disable (DEACTIVATED) a connection.
Update an existing connection's name, description, or warehouse configuration (e.g. register new tables/schemas). Validates against the warehouse by default.
Error handling lacks recovery guidance. No tool descriptions explain what to do on failure (e.g., 'If connection validation fails, check credentials and retry' or 'If metadata not found, call search_metadata first').
Destructive operations (delete_connection, delete_metadata) lack confirmation/dry-run support. Agents can permanently delete resources without a safety gate.
Export tools (export_liveboard_report, export_answer_report) write to local filesystem via output_path parameter. No validation of path traversal, no guidance on where files are written or how to retrieve them.