Loading ToolBench
MCP server that wraps WolframScript for evaluating Wolfram Language scripts and querying Wolfram Engine information
Server has 4 tools with basic descriptions but significant quality gaps. Tool naming is reasonable (evaluate, version_wolframscript, version_wolframengine, licensetype) but descriptions lack depth and parameter documentation is minimal. The 'evaluate' tool has a detailed description with examples, but other tools are sparse (10-20 chars). No input schemas visible for version/license tools, they accept empty objects but lack formal schema documentation. No output schemas documented for any tool. Error handling is present in the implementation but not surfaced in tool descriptions. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only.
Evaluate a Wolfram Language script by running the `wolframscript -print -file <script>` command. Documentation for Wolfram Language: https://context7.com/websites/reference_wolfram_language/llms.txt IMPORTANT: The tool is returning the result of the last line executed in the script, and any expression printed explicitly with `Print[]`.
Get the license type of the Wolfram Engine.
Get the version of the Wolfram Engine.
Get the version of the `wolframscript` tool.
Three tools (version_wolframscript, version_wolframengine, licensetype) have descriptions under 20 characters ('Get the version of the `wolframscript` tool.' = 47 chars, but 'Get the version of the Wolfram Engine.' = 38 chars, 'Get the license type of the Wolfram Engine.' = 42 chars). These are borderline but lack context on when/why to call them.
No input schemas visible for version_wolframscript, version_wolframengine, licensetype. They accept empty objects but no formal JSON Schema is documented.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract fields without knowing return structure. 'evaluate' returns a string, but version/license tools' output format is undocumented.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 40 | <=2025-11-25 | v2 |
All tools are read-only but lack tool annotations (readOnlyHint). This prevents clients from filtering safe vs. destructive operations and optimizing caching.
Error handling is implemented in _run_wolframscript (RuntimeError on command failure, FileNotFoundError for missing wolframscript) but not surfaced in tool descriptions. LLMs cannot anticipate failure modes or recovery steps.