Jupyter notebook MCP server for Antigravity — edit, run, and orchestrate notebooks via AI agents
Server has 10 tools with complete input schemas and descriptions present for all tools and most parameters. However, descriptions are generic and lack LLM-optimization guidance (e.g., 'when to use', 'prerequisites', 'what happens'). Parameter descriptions are sparse or missing for optional parameters (save, checkpoint, kernel_name). No output schemas documented. Error handling guidance absent. Tool names follow verb_noun convention well (open_, list_, read_, edit_, insert_, delete_, run_, restart_). Composition is reasonable, tools are single-purpose and chainable. Security concerns: no mention of path traversal validation for notebook_path parameter, which accepts arbitrary strings.
Delete the cell at the given index. Creates a timestamped checkpoint by default.
Replace the source of a cell. Creates a timestamped checkpoint by default.
Insert a new cell at the given index, shifting later cells down.
List all cells in a notebook with their index, type, tags, and an optional source preview.
Open a notebook and return its metadata (cell count, pipeline stages).
Read the full source, outputs, and metadata of a single cell.
Restart the Jupyter kernel for a notebook, clearing all in-memory state.
Tool descriptions lack LLM-optimization guidance. Descriptions state WHAT but omit WHEN to use, prerequisites, and consequences. E.g., 'Execute a single code cell' does not explain when to call run_cell vs run_range, or that execution has side effects.
Optional parameters (save, checkpoint, kernel_name, timeout, stop_on_error) lack descriptions in schema. LLMs cannot infer parameter meaning from names alone. E.g., 'save' could mean save to disk, save to database, or save state.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., cell_id, execution_result, error_message). Responses are unstructured.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Execute a single code cell and return its outputs.
Execute cells grouped by their tag-based pipeline stages.
Execute all code cells in the inclusive range [start, end].
No error handling guidance. Descriptions do not explain what errors are possible, whether they are retryable, or what the LLM should do next. E.g., 'notebook not found' vs 'kernel timeout' require different recovery strategies.
notebook_path parameter accepts arbitrary strings with no validation guidance. No mention of path traversal protection, symlink handling, or workspace boundary enforcement. LLMs could be tricked into accessing files outside intended scope.