A reproducible, zero-prerequisite Jupyter notebook environment with MCP agent interface, running in Docker or locally with uv-managed Python venvs
Cellar provides 17 tools with consistent naming (verb_noun pattern: get_, add_, run_, set_, delete_, move_) and comprehensive descriptions (avg ~120 chars). All tools have input schemas with type definitions and parameter descriptions. However, output schemas are not documented in the source, parameter descriptions lack format/constraint details (e.g., no regex patterns, min/max for numeric fields), and error handling guidance is minimal. Tool composition is strong (single responsibility, chainable IDs), but descriptions could be more LLM-optimized per the 50-200 char baseline. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels being tracked.
Add a cell and run it immediately, waiting for completion with optional progress notifications
Add a single cell to the notebook, optionally after a specified cell, with optional type, source, and metadata
Add multiple cells to the notebook in a single batch operation
Delete a cell from the notebook
Get a cell's full content: source, outputs, and metadata. Resolves any cell reference (short handle, prefix, or full UUID)
Get the current notebook's cell map: a flat list of all cells with their ids, types, and source line ranges, for agent navigation and cell reference resolution
Search for files matching a glob pattern in the workspace
Output schemas not documented. Tools return results but LLMs cannot plan downstream calls without knowing response structure (e.g., what fields does get_cell return? Is there pagination?). Violates pattern:tool requirement.
Parameter descriptions lack format constraints. 'notebook' param described as 'workspace-relative notebook path' but no guidance on format (e.g., must it end in .ipynb? Can it include ../ for traversal?). 'cellType' enum is present but other params lack min/max, regex, or length bounds. Violates pattern:constrained-input.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | <=2025-11-25 | v2 |
Search for text patterns in files within the workspace
Move a cell to a new position in the notebook
Read a file from the workspace
Run an existing cell and wait for completion, with optional progress notifications for long-running cells
Set a cell's export target (module name and line range for code generation)
Replace a cell's source code
Change a cell's type (code, markdown, sql, raw, chat)
Set the export target for a notebook (output module path and language)
Search the web for information
Write content to a file in the workspace
No tool annotations despite clear risk levels. Tools marked DESTRUCTIVE (delete_cell) or WRITE (add_cell, run_cell, write) lack readOnlyHint/destructiveHint/idempotentHint annotations. LLMs cannot distinguish safe from risky tools without explicit hints. Violates current MCP spec (2026-07-28) tool annotation requirement.
Error handling guidance missing. No descriptions of what errors tools return, how to recover, or what the LLM should do next. E.g., delete_cell could fail if cell not found, but no guidance on retry vs user correction. Violates pattern:recovery-guide.
Batch tool (add_cells) lacks pagination/limit guidance. No indication of max cells per batch, whether results are paginated, or how to handle large operations. Violates pattern:paginated-result for list-returning tools.