Loading ToolBench
MCP server that returns the current directory path
Single tool with minimal definition quality. Tool name 'get_current_directory' follows verb_noun convention (+), but description is only 79 chars and lacks context on WHEN to use it or what downstream tools might need. Input schema is empty object {} with no parameters, technically valid but offers no guidance. Output is unstructured JSON text (not typed fields), forcing LLM to parse. No error handling, no parameter descriptions (none exist), no output schema documentation. Lacks composition guidance and chaining IDs. This is a toy tool with no production-grade polish.
Returns the current directory path where the MCP server is running
No output schema documented. Tool returns unstructured JSON text wrapped in content array, forcing LLM to parse. Should declare typed fields (currentDirectory: string, platform: string, nodeVersion: string).
Description is 79 characters but lacks context. Does not explain WHEN to call this tool, what downstream tools might need the directory, or whether it reflects the agent's working directory or server's. LLM cannot infer intent.
Empty input schema {} provides no guidance. Even though no parameters are needed, the schema should be explicit: {"type": "object", "properties": {}, "required": []}. Current form is ambiguous.
No error handling. Tool does not document what happens if process.cwd() fails, or how LLM should recover. No recovery guidance provided.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Tool returns platform and nodeVersion alongside currentDirectory, but no downstream tool is documented that consumes these fields. Violates response-shaper pattern, return only what the agent needs.