An MCP server that executes Python code in isolated sandbox environments with support for file mounting, artifact generation, and session persistence.
Server has 6 tools with basic descriptions and some schema definition, but significant gaps in parameter documentation and output schema clarity. Tool names are action-oriented (run_code, persist_artifact, workspace_inspect, list_dir, preview_file, mount_file), which is good. However, most parameter descriptions are minimal (1-2 lines), and output schemas are not explicitly documented in the visible code. The run_code tool accepts complex nested objects (files array with url/mountPath) but lacks detailed constraints. Error handling is minimal, no recovery guidance or categorization visible. Security concerns exist: run_code executes arbitrary Python, and artifact serving relies on session headers without clear validation. Composition is reasonable (tools are single-purpose), but parameter naming could be more explicit (e.g., 'dir_path' vs 'path').
List contents of a directory in the session workspace.
Mount a remote file into the session workspace for use by subsequent code executions.
Persist an artifact file for the current session.
Preview the contents of a text file in the workspace.
Execute Python code inside an isolated virtual-env and return captured output. Artifacts are returned as paths relative to the output directory. Only files inside output/ are included.
Inspect the workspace directory structure and file contents.
persist_artifact and workspace_inspect lack visible input schemas and output documentation. Cannot verify parameter types or return structure.
persist_artifact has only 2-word description ('Persist an artifact file for the current session'). Lacks WHAT it does, WHEN to use it, and what it returns. Below 20-char minimum for actionable guidance.
run_code tool accepts 'requirements' (pip packages) and 'files' (remote downloads) but lacks constraints: no max package count, no URL validation, no file size limits. Unbounded inputs risk timeouts or resource exhaustion.
No error handling guidance visible. Tools do not document what errors are retryable, user-fixable, or fatal. LLM cannot determine recovery strategy.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
run_code executes arbitrary Python code with no dry-run or confirmation step. Irreversible operations (file deletion, external API calls) lack safeguards against agent mistakes.