Loading ToolBench
Model Context Protocol (MCP) server for Teradata, providing tools for database operations, SQL execution, backup/restore, feature store access, vector store operations, and graph analysis
The Teradata MCP server demonstrates uneven quality across 19 tools. Strengths: clear verb-noun naming conventions (base_readQuery, fs_getDataDomains, graph_traceLineage), detailed descriptions for most tools explaining WHEN to use them and distinguishing between similar tools (e.g., base_readQuery vs base_tablePreview). Weaknesses: input parameter descriptions are largely absent or minimal; most tools lack documented output schemas; no visible error handling guidance or recovery patterns; several tools accept object parameters (fs_config) with no inline schema definition; security posture unclear (no evidence of credential injection or permission gating); tools like util_base_dynamicQuery accept callable parameters, impossible to validate or reason about. The server follows HTTP/Streamable HTTP transport correctly but lacks modern MCP patterns like tool annotations (readOnlyHint, destructiveHint) and structured error responses.
Execute a user-provided SQL query against Teradata and return the results. Use this tool ONLY when the user supplies an explicit SQL statement or a request that includes filter conditions (WHERE clause, aggregations, JOINs, etc.). Do NOT use for simply browsing or sampling rows from a table — use base_tablePreview for that. The sql parameter is required and must contain the full SQL text.
Extract the DDL for a Teradata table, view, or stored procedure and SAVE it as a .sql file on disk. Use this tool ONLY when the user explicitly wants to export, write, download, or persist DDL to a file. Do NOT use simply to display or view DDL in the conversation — use base_tableDDL to display DDL without saving.
Execute chat completion using OpenAI-compatible LLM inference server. Calls the Teradata CompleteChat table operator to generate text completions, summaries, classifications, and other AI-powered text transformations.
Configure a disk file system for DSA backup operations. Adds a new disk file system to the existing list or updates an existing one. This allows DSA to use the file system for backup storage operations.
Unspecified object parameters (fs_config) lack inline schema definition. LLM cannot determine what fields are required or optional.
No documented return schemas for any tool. LLM cannot plan downstream tool calls or know what fields to extract.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 69 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 32 | - | v1 |
Delete all disk file system configurations from DSA. Removes all disk file system configurations from DSA. This operation will fail if any file systems are currently in use by backup operations or file target groups.
Returns a summary of the feature store content. Use this to understand what data is available in the feature store.
List the available entities for a given data domain. Requires a configured `database_name` and `data_domain` and `entity` in the feature store config. Use this to explore which entities can be used when building a dataset.
List the available data domains. Requires a configured `database_name` in the feature store config. Use this to explore which entities can be used when building a dataset.
Returns the feature store data model, including the feature catalog, process catalog, and dataset catalog.
Check if a feature store is present in the specified database.
Comprehensive single-fetch analysis of database dependencies including roots, cycles, components, and blast radius.
Perform breadth-first search (BFS) on a dependency graph to determine distance levels from seed objects.
Partition a graph into connected components using Union-Find algorithm.
Detect cycles in a dependency graph using Union-Find and iterative DFS algorithms.
Generate DDL for creating an edge repository table used by graph analysis tools.
Discover root objects in a dependency graph - objects with no upstream dependencies.
Trace lineage and dependencies in a graph using recursive CTEs to identify impact paths and reachable subgraphs.
List all configured disk file systems in DSA. Lists all disk file systems configured for backup operations, showing file system paths, maximum files allowed per file system, and configuration status.
This tool is used to execute dynamic SQL queries that are generated at runtime by a generator function.
Callable parameter (sql_generator in util_base_dynamicQuery) cannot be represented in JSON Schema and is unmappable by LLMs. Violates composability principle.
Destructive tool (delete_disk_file_systems) lacks confirmation, dry-run, or explicit user input requirement. No protection against accidental execution.
No error handling guidance across any tool. Descriptions lack recovery hints like 'If user not found, call search_users()' or error categorization (retryable vs fatal).
Format and constraint specifications missing from many parameters. 'edge_repository' (graph tools), 'direction', and 'object_type' lack enum constraints or format documentation.
Pagination and result limits not documented. Tools like list_disk_file_systems, graph_findRootObjects, and graph_analyseDatabase may return unbounded results, risking context window exhaustion.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) visible in definitions. LLM cannot infer which tools are safe to retry or have irreversible consequences.
Credential and permission handling not documented. No evidence of server-side secret injection for database credentials or permission gating for destructive operations.