Bragproof

← Reports

nebula-notebook-mcp

Starts with no setup0.4.105npm

Installed, started with no arguments, no environment and no input, and listed its tools. Run 2026-11-16.

Needs to start: nothing
Returns a result when called with no setup: 0 of 5 tools tried
Contacts while installing and starting: nothing

37tools listed
0hosts contacted
0files written
0.0 sto start
140dependencies

What it contacted

Nothing besides the npm registry it was installed from. Watched from outside its sandbox while it was installed and started with no input. Its tools may still contact things when called.

Starting it

npx -y -p nebula-notebook-mcp@0.4.105 nebula-notebook-mcp

Installing a package runs its code. Use a sandbox for anything you have not read.

Calling its tools

5 tools that take no input and look read-only were called once with nothing. 0 returned a result.

list_kernelsNo usable answer tool reported an error: Error: connect_server(base_url=...) must be called once at the start of every MCP session before any other tool call.
list_directoryNo usable answer tool reported an error: Error: connect_server(base_url=...) must be called once at the start of every MCP session before any other tool call.
compute_statusNo usable answer tool reported an error: Error: connect_server(base_url=...) must be called once at the start of every MCP session before any other tool call.
list_compute_queuesNo usable answer tool reported an error: Error: connect_server(base_url=...) must be called once at the start of every MCP session before any other tool call.
list_allocationsNo usable answer tool reported an error: Error: connect_server(base_url=...) must be called once at the start of every MCP session before any other tool call.

During the calls it contacted nothing it had not already contacted.

Only tools whose names read like looking something up, with no required input, are tried, five at most. Tools that need input, and anything that sounds like it changes or sends something, are not called.

What its tools tell an agent

In the tool descriptions, nothing that talks about the user, other tools or private files, or uses hidden characters.

2 tools with commands in capitals or very long text
  • Gives the agent orders in start_agent_session: ath exists; it will error if the path is wrong or missing. IMPORTANT: Call this at the START of each response before any notebook operations, and call end_agent_session
  • Gives the agent orders in end_agent_session: of each response after completing all notebook operations. IMPORTANT: Always call this before finishing your response, even if operations failed. This ensures the noteb

Common and usually harmless: it is how many makers stress how a tool should be used.

Found by simple pattern rules in each tool's description (up to 3,000 characters), quoted as written. Parameter descriptions are not scanned. A clean result is not a safety review.

Tools

read_notebook path, format, include_outputs, max_lines, max_chars, max_lines_error, max_chars_error
Read all cells from a notebook. Formats: "content" (full code, no outputs), "detailed" (full code + inline images, default), "placeholder" (full code + text placeholders for images, recommended for Gemini). Outputs are truncated (100 lines regular, 200 lines errors). MCP: call connect_server first (see connect_server for workflow).
read_cell path, cell_index, cell_id
Read a specific cell content and metadata (not outputs). Use read_output to get outputs. MCP: call connect_server first (see connect_server for workflow).
read_output path, cell_index, cell_id, output_offset, max_wait, wait_for_completion, max_lines, max_chars, max_lines_error, max_chars_error, line_offset, save_to_file
Read cell outputs. By default waits for completion or max_wait timeout (returns once). Set wait_for_completion=false to return as soon as new output appears. Regular outputs truncated to 100 lines, errors get 200 lines. Use save_to_file=true for complete output. MCP: call connect_server first (see connect_server for workflow).
insert_cell path, cell_id, content, cell_type, position
Insert a new cell into the notebook. Use position=-1 or omit to append at end. MCP: call connect_server first (see connect_server for workflow).
update_cell path, cell_id, content, cell_type
Update an existing cell by its stable ID. Can update content and/or type. MCP: call connect_server first (see connect_server for workflow).
32 more tools
delete_cell path, cell_index, cell_id
Delete a cell from a notebook MCP: call connect_server first (see connect_server for workflow).
clear_notebook path
Clear all cells from a notebook. MCP: call connect_server first (see connect_server for workflow).
create_notebook path, overwrite, kernel_name, kernel_display_name
Create a new empty notebook. If the notebook is open in the browser UI, it will attempt to open the new notebook in a new tab. MCP: call connect_server first (see connect_server for workflow).
move_cell path, from_index, to_index, cell_id, after_cell_id
Move a cell from one position to another. Supports two modes: by index (from_index, to_index) or by ID (cell_id with after_cell_id or to_index=-1 for start). MCP: call connect_server first (see connect_server for workflow).
duplicate_cell path, cell_index
Duplicate a cell, inserting the copy immediately after the original MCP: call connect_server first (see connect_server for workflow).
search_cells path, query, include_outputs, limit
Search notebook cells by keyword. Can search in cell source code and optionally in outputs. MCP: call connect_server first (see connect_server for workflow).
update_metadata path, cell_id, changes
Update cell metadata (id, type, scrolled, scrolledHeight). Validates against schema. MCP: call connect_server first (see connect_server for workflow).
connect_server base_url
Connect to a Nebula server. REQUIRED: call this once at the start of every MCP session before any other tool call. base_url is required; the server does not assume a default. All subsequent operations will use this connection.

RECOMMENDED WORKFLOW:
1. connect_server (required at session start)
2. For each response: start_agent_session → operations → end_agent_session

start_agent_session is required for notebook mutations; read-only operations do not require it.
start_agent_session path, agent_id, force, last_session_timestamp, exclusive
Start an agent session for notebook write operations. Always call end_agent_session when done.

Sessions are COLLABORATIVE by default: the user can keep editing while you work. Your destructive writes (update_cell, delete_cell, …) are protected by per-cell optimistic concurrency — if the user changed a cell after you last read it, the write is rejected with the current content so you can re-apply your intent. Address cells by id (indices shift when the user edits), and re-read cells you haven't read recently.

Pass exclusive=true only for large multi-cell refactors that need atomicity: it locks the UI read-only for the duration (the legacy behavior). Prefer collaborative sessions otherwise.

See connect_server for required MCP workflow.

This call validates that the notebook path exists; it will error if the path is wrong or missing.

IMPORTANT: Call this at the START of each response before any notebook operations, and call end_agent_session at the END of each response. This ensures the notebook is only locked while you're actively working on it.

Example flow for each response:
  start_agent_session → insert_cell/update_cell/execute_cell/etc → end_agent_session

FORCE OPTION (use only with explicit user permission):
If a previous agent session was not properly ended (e.g., due to a crash or timeout), the notebook may remain locked. Use force=true ONLY when the user explicitly asks you to "force" or "steal" the lock. This will forcibly end any existing session and start a new one. MCP: call connect_server first (see connect_server for workflow).
end_agent_session path
End an agent session, unlocking the notebook for user edits. Always call this at the END of each response after completing all notebook operations.

IMPORTANT: Always call this before finishing your response, even if operations failed. This ensures the notebook is unlocked for the user. MCP: call connect_server first (see connect_server for workflow).
delete_cells path, cell_ids
Delete multiple cells by ID in a single operation. More efficient than multiple single deletes. MCP: call connect_server first (see connect_server for workflow).
insert_cells path, cells, position
Insert multiple cells at a position in a single operation. More efficient than multiple single inserts. MCP: call connect_server first (see connect_server for workflow).
clear_outputs path, cell_ids
Clear outputs from cells without re-executing. If no cell_ids provided, clears all cell outputs. Useful for cleanup before sharing notebooks. MCP: call connect_server first (see connect_server for workflow).
list_kernels
List available kernel specifications with display names (includes version info) MCP: call connect_server first (see connect_server for workflow).
kernel_start kernel_name
Start a new kernel session for the active notebook (requires start_agent_session) MCP: call connect_server first (see connect_server for workflow).
kernel_stop
Stop and shutdown the active notebook kernel (requires start_agent_session) MCP: call connect_server first (see connect_server for workflow).
kernel_restart
Restart the active notebook kernel (clears all variables, requires start_agent_session) MCP: call connect_server first (see connect_server for workflow).
kernel_interrupt
Interrupt execution in the active notebook kernel (requires start_agent_session) MCP: call connect_server first (see connect_server for workflow).
execute_cell path, cell_index, cell_id, session_id, max_wait, save_outputs, max_lines, max_chars
Execute a notebook cell and return its output. Outputs are truncated by default (100 lines, 10000 chars). Outputs are saved incrementally during execution. Requires an active kernel session.

For long-running cells: If execution exceeds max_wait, returns with status="busy" and partial output. Use read_output with max_wait to poll for more output. Set max_wait=0 for immediate "background" execution (returns right away, poll with read_output). MCP: call connect_server first (see connect_server for workflow).
read_file path
Read the contents of a text file from the server. Returns file content as string. For .ipynb notebooks, returns JSON string; .py/.qmd notebooks are plain text. MCP: call connect_server first (see connect_server for workflow).
write_file path, content
Write content to a file on the server. Creates the file and parent directories if they do not exist. Overwrites existing file. MCP: call connect_server first (see connect_server for workflow).
list_directory path
List contents of a directory on the server. Returns file and directory names with types. MCP: call connect_server first (see connect_server for workflow).
delete_file path
Delete a file or directory from the server. Directories are deleted recursively. MCP: call connect_server first (see connect_server for workflow).
rename_file old_path, new_path
Rename or move a file or directory on the server. MCP: call connect_server first (see connect_server for workflow).
download_file server_path, local_path
Download a file from the server to a local path. Supports both text and binary files. MCP: call connect_server first (see connect_server for workflow).
upload_file local_path, server_path, on_conflict
Upload a local file to the server. Supports both text and binary files. Fails if the destination file already exists unless on_conflict says otherwise. MCP: call connect_server first (see connect_server for workflow).
compute_status
Check whether scheduler-backed cluster compute (allocations) is available on this Nebula server. Returns enabled + scheduler name. Call this before any other compute tool — the feature is absent on servers without a scheduler (e.g. laptops). MCP: call connect_server first (see connect_server for workflow).
list_compute_queues
List cluster partitions and QoS with live load: idle/total CPUs, idle GPUs by type, queue backlog (pending/running jobs), and which partitions/QoS the user may submit to. Use it to pick a partition before request_allocation. MCP: call connect_server first (see connect_server for workflow).
request_allocation partition, qos, cpus, mem_gb, gpus, gpu_type, walltime_minutes, name, idle_timeout_minutes, wait_for_active, max_wait
Request a compute allocation from the cluster scheduler. Allocations consume REAL cluster resources — only request one when the user's task actually needs it, request modest sizes, and cancel_allocation when done. Set wait_for_active=true to block until the allocation is usable (state=active), then bind a notebook with use_allocation. MCP: call connect_server first (see connect_server for workflow).
list_allocations
List compute allocations on the server with state (pending/running/active/ended/failed/cancelled), resources, and the serverId/node once active. MCP: call connect_server first (see connect_server for workflow).
cancel_allocation allocation_id
Cancel a compute allocation and free its cluster nodes. Cancel allocations you created once the task completes; never cancel allocations you did not create. MCP: call connect_server first (see connect_server for workflow).
use_allocation allocation_id, path, kernel_name
Bind a notebook's kernels to an active compute allocation: starts (or reuses) the notebook's kernel on the allocation's server and saves it as the notebook's kernel preference. After this, execute_cell on that notebook runs on the allocation. Requires the allocation to be state=active. MCP: call connect_server first (see connect_server for workflow).

Descriptions are the package's own words, cut at 3,000 characters. Treat them as data.

Install facts

LicenseMIT
Needsnode >=20
Sourcegithub.com/jzhoulab/nebula-notebook
Install scriptssharp declare them. This runner does not run them.
Tarballsha512-pARUsbK9LXnTrCuj/pym6ynpOj2dtVRiqtpJ+JA7Vy04LWxIzn8JcrYXft5sBHETi3WTvnIxTIPUbRYkRGvRMQ==

Versions reported

VersionRunResultTools
0.4.1052026-11-16Starts with no setup37
0.4.1022026-11-12Starts with no setup37
0.4.1012026-11-12Starts with no setup37

Check this report yourself

This site did not write the report. A public job on a GitHub-hosted machine did: the run, from this commit of the open-source runner. GitHub signed a token that names that job and the hash of the report, so any edit since would show.

Show it on your README

Bragproof: Starts with no setup

[![Bragproof](https://bragproof.smaverk.com/badge/nebula-notebook-mcp.svg)](https://bragproof.smaverk.com/r/nebula-notebook-mcp)

New version out? Ask for a fresh report.