Loading ToolBench
pwalker-mcp provides 6 tools with basic but incomplete definitions. All tools have names and descriptions, and most have input schemas defined via Zod. However, the server has significant gaps in description quality, parameter documentation, output schema documentation, and error handling guidance. Descriptions are functional but brief (averaging ~80 chars), and parameter descriptions lack detail on constraints, formats, and expected values. Output structures are returned as plain text rather than documented structured objects. No tool annotations (readOnlyHint/destructiveHint/idempotentHint) are present. Error messages are generic and do not guide recovery. The task queue and process management tools are reasonably well-scoped, but lack the LLM-facing optimization expected in production agent tools.
Get the stored stdout and stderr output for a process with the given ID. Also returns the exit code of the process if it has exited.
Kill a process with the given ID.
Launch a long-running background process with the given command and arguments. This is useful for processes that run continuously and watch for file changes.
Pop a task from the task queue. Returns 'No tasks in the queue.' if the queue is empty.
Push a list of tasks and/or tasks from a taskfile to the task queue. Tasks can be retrieved using the 'pop-task' tool.
Wait for a specified number of seconds before continuing.
Output schemas not documented. Tools return plain text wrapped in content arrays, but no structured schema is declared. LLMs cannot determine what fields to expect or plan downstream tool chaining.
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint attributes. LLMs cannot determine which tools are safe to retry or have side effects.
Parameter descriptions lack constraint details. 'seconds' parameter has min=0 in schema but description does not explain upper bounds, reasonable ranges, or impact of very large values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 54 | - | v1 |
Error responses are generic and do not guide recovery. 'Process not found' and 'No tasks in the queue' do not suggest alternatives or next steps for the LLM.
Process ID return value is not documented. launch-background-process returns a UUID string, but no description explains the format or how to use it with get-process-output and kill-process.
pop-task returns plain text task string with no metadata. If multiple tasks have been queued, the LLM cannot distinguish task ID, task status, or priority from the returned text.
Parameter taskfile documented as 'absolute path' but no guidance on path traversal constraints or validation. Agents could be tricked into reading unauthorized files.
launch-background-process does not document timeout behavior, maximum process lifetime, or resource limits. An agent could spawn unlimited long-running processes, exhausting system resources.