MCP server that executes JavaScript code in an isolated VM with plugin access
This server has significant definition quality gaps. Tool naming is non-standard (uses hyphens instead of verb_noun convention). Descriptions lack depth and contain profanity ('list some shit'). The 'run-it' tool exposes arbitrary code execution without proper input validation guidance, parameter constraints, or output schema documentation. The 'list-it' tool has a vague, unprofessional description. No error handling guidance, no output schemas documented, no parameter constraints (enums, formats, ranges). Parameter descriptions are minimal. No security considerations for the dangerous 'run-it' tool. Schemas are present but lack richness, no examples beyond one in 'run-it', no descriptions of return structure.
list some shit
execute javascript code in an isolated vm with plugin access. your code is wrapped in `async function run() { ... }` and the result is returned. use 'return' to return values from your code.
Tool names do not follow verb_noun convention. 'run-it' and 'list-it' use hyphens and lack clarity. Should be 'run_code' or 'execute_code' and 'list_plugins'.
Description for 'list-it' is unprofessional and vague: 'list some shit'. Does not explain what is being listed, when to use it, or what structure is returned. Violates pattern:tool-description baseline (all tools require meaningful 10-1024 char descriptions).
The 'run-it' tool allows arbitrary JavaScript execution in a VM. This is an IRREVERSIBLE, HIGH-RISK operation but has no output schema documentation, no error guidance, no validation of code safety, and no description of what constitutes valid input. Agents cannot plan around this tool safely.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 26 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 27 | - | v1 |
No output schemas documented. The tool definitions do not specify what fields, types, or structure the LLM should expect from responses.
Parameter 'pluginName' in 'list-it' lacks a type declaration in the schema. The description says 'optional' but JSON Schema does not mark it as non-required or provide a format/pattern. Schema should include minLength, maxLength, or a pattern constraint to guide the LLM.
The 'code' parameter in 'run-it' has no length limit, timeout guidance, or security constraints. An LLM could pass arbitrarily long, malicious, or infinite-loop code. Description should include: max code length, timeout behavior, and what sandbox restrictions apply.
No error handling or recovery guidance. If code execution fails (timeout, syntax error, runtime error), the LLM receives only a generic 'failed to execute code' message with the error message. Does not explain whether the call is retryable, how to fix the input, or when to give up.
No mention of idempotency. The 'run-it' tool returns the result of arbitrary code execution, if called twice with the same code, results may differ (e.g. if code reads from files, calls external APIs, or uses random numbers). Agents need to know whether retrying will produce the same outcome.
Tool composition is minimal. No guidance on how to chain these tools together. E.g., does 'list-it' return plugin names/IDs that can be passed to 'run-it'? Is there a standard way to invoke a plugin method via 'run-it' after discovering it with 'list-it'?