MCP server that wraps the Dagster GraphQL API — manage runs, assets, schedules, sensors, and backfills from any MCP client
Nothing to run from here
Colab-MCP (unofficial) is installed from its publisher's own source and answers where it runs, so this marketplace is not in the path of a single call. There is no address here to send one to, and a panel that pretended otherwise would be showing you an answer we made up. Install it and call it from your own client — the Installation tab has the entry for each one.
90/100
trust score
50
tools declared
local
runs on your machine
—
downloads a day
What it does
Unofficial MCP server for Google Colab: runtimes, code, long jobs, files and notebooks.
Quickstart
# 1 — add it to Claude Code
claude mcp add --scope user colab-mcp-unofficial '--' uvx colab-cli-mcp
# 2 — ask your agent for something these tools do
# colab_sessions, colab_status, colab_usage and 47 more
Colab-MCP (unofficial) costs nothing on mcprush: there is no plan to choose here, no cap to set and nothing mcprush can bill you for.
Publishernot claimed
AMAll 1 server
amgallego81 server
1 category
Collected from a public index. Nobody has claimed this account, so nothing here was written by its author — claim it if it is yours.
Installation
Where are you running it?
Every route below installs the same thing and ends at the same approval screen. This one runs on your machine: your client starts Colab-MCP (unofficial) as a process under your own user, with your files and your network, so the tool surface below is what it can do to your computer rather than to a server somewhere else. Its scan result is on the Security tab — read the surface before you approve it.
Claude Codedeclared by this listing
This is a public server: you run it yourself and this marketplace is not in the path. You need uv (docs.astral.sh/uv): uvx comes with it. Claude Code registers it in one command; --scope user makes it available in every project. The '--' is in quotes so that PowerShell passes it on.
claude mcp add --scope user colab-mcp-unofficial '--' uvx colab-cli-mcpReconnect, or start a new session, and the tools appear in the model’s tool list.
Free on mcprush, and it runs from its publisher’s own package: this marketplace is not in the path, so there is no key of ours to hold and no card to add.
What it asks of you
Nothing is required, but it can take COLAB_MCP_WORKDIR (The only local folder the server reads and writes. Defaults to the current directory.), COLAB_MCP_DRIVE (Set to 1 to enable the optional Google Drive tools (they need a one-time sign-in, see the README).) if you want to set them.
What it writes
One config entry your client uses to start the process locally. A local server runs with your file system and your network, which is why it is priced without metering.
What you approve
50 tools, with what each one reads, writes and reaches shown before you agree — the same list on every route above. Read the tool surface.
Tool surface
What the model actually sees. Descriptions are diffed on every release — see version history.
List active Colab runtimes (sessions). Also a quick check that Google sign-in works.
No parameter schema on file.
Show hardware, machine shape and state of a session (or the only active one).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
Show the account's compute-unit balance and current burn rate.
No parameter schema on file.
Create a runtime. CPU by default; gpu: T4, L4, G4, H100 or A100; tpu: v5e1 or v6e1; high_mem needs Colab Pro. Consumes compute units until stopped.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
| gpu | string | no | — |
| tpu | string | no | — |
| high_mem | boolean | no | — |
Make sure a named session exists: reuse it if colab status finds it, otherwise create it. Safe to call repeatedly.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | yes | — |
| gpu | string | no | — |
| tpu | string | no | — |
| high_mem | boolean | no | — |
Restart the Jupyter kernel of a session: clears all Python variables but keeps the VM and its files.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
Print the browser URL that attaches the Colab UI to this session (does not open a browser).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
Report the VM's Python version, CPUs, RAM, free disk and GPU (if any).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
Stop a session and release the VM (stops consuming compute units). Files not downloaded are lost.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
Install Python packages on the runtime (uses uv, falls back to pip). Names/specifiers only, e.g. 'qiskit', 'pandas>=2.0'.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| packages | array | yes | — |
| session | string | no | — |
Run Python on the runtime; variables persist between calls (same kernel). Give code (text) or file (local .py/.ipynb inside the working directory), not both.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| code | string | no | — |
| file | string | no | — |
| session | string | no | — |
| timeout | number | no | — |
| secrets | array | no | — |
| plot_path | string | no | — |
One-shot job: allocate a fresh VM, run a local script (inside the working directory) with args, then release the VM (or keep it with keep=true).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| script | string | yes | — |
| args | array | no | — |
| gpu | string | no | — |
| high_mem | boolean | no | — |
| keep | boolean | no | — |
| timeout | number | no | — |
Start a shell command in the background on the VM (e.g. 'python -u train.py --epochs 50') and return a job id. Use for anything longer than a few minutes; poll with colab_job_status / colab_job_logs. Jobs die if the VM is stopped.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| command | string | yes | — |
| name | string | no | — |
| session | string | no | — |
State of a background job: running, finished (with exit code), cancelled, or lost.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| job_id | string | yes | — |
| session | string | no | — |
Last tail lines of a background job's combined stdout/stderr, plus its state.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| job_id | string | yes | — |
| tail | integer | no | — |
| session | string | no | — |
Terminate a running background job (SIGTERM to its process group).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| job_id | string | yes | — |
| session | string | no | — |
List all background jobs started on this VM.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
Create a local .ipynb inside the working directory. cells: list of {"type": "code"|"markdown", "source": "..."}.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| cells | array | no | — |
| overwrite | boolean | no | — |
List .ipynb files under a directory of the working directory.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| directory | string | no | — |
Show a notebook's cells (index, id, type, source) and, optionally, their saved outputs. Use start/end to page through big notebooks.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| include_outputs | boolean | no | — |
| start | integer | no | — |
| end | integer | no | — |
Insert a cell. Default is at the end; or give position (index) or after (cell id/index). cell_type: code or markdown.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| source | string | yes | — |
| cell_type | string | no | — |
| position | integer | no | — |
| after | string | no | — |
Replace a cell's source (and optionally its type). cell is an id or an index. Clears that cell's stale outputs.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| cell | string | yes | — |
| source | string | yes | — |
| cell_type | string | no | — |
Delete one cell by id or index.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| cell | string | yes | — |
Move a cell to a new index.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| cell | string | yes | — |
| to_index | integer | yes | — |
Clear saved outputs of one cell, or of all cells.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| cell | string | no | — |
Run ONE code cell on the runtime (state persists between cells) and store its output in the notebook. Reports ok/ERROR status. A displayed plot is saved under .colab-mcp/images/.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| cell | string | yes | — |
| session | string | no | — |
| timeout | number | no | — |
Run every code cell in order (from index start), storing outputs in the notebook. Stops at the first failing cell unless stop_on_error=false.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| session | string | no | — |
| timeout_per_cell | number | no | — |
| stop_on_error | boolean | no | — |
| start | integer | no | — |
Convert a notebook to a .py script (markdown becomes comments; lines starting with ! or % are commented out). With out_path, writes the file; otherwise returns the text. Handy before colab_run.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| out_path | string | no | — |
Delete a local notebook by moving it to .colab-mcp-trash/ inside the working directory (recoverable, never a hard delete).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
Upload one local file (inside the working directory) to the VM, e.g. remote_path '/content/data.csv'. Limited by COLAB_MCP_MAX_UPLOAD_MB (default 100): for big data, download it on the VM instead (colab_exec / colab_job_start).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| local_path | string | yes | — |
| remote_path | string | yes | — |
| session | string | no | — |
Download one file from the VM into the working directory.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| remote_path | string | yes | — |
| local_path | string | yes | — |
| session | string | no | — |
Upload a whole local folder (inside the working directory) to the VM as one zip, extracted at remote_dir. Skips .git, caches, symlinks and likely secret files (.env, *.key, *.pem...).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| local_dir | string | yes | — |
| remote_dir | string | yes | — |
| session | string | no | — |
Download a whole VM folder into local_dir (inside the working directory) via one zip. Refuses archives whose paths escape the target or exceed COLAB_MCP_MAX_DOWNLOAD_MB (default 1024).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| remote_dir | string | yes | — |
| local_dir | string | yes | — |
| session | string | no | — |
List files in a VM directory (default /content).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | no | — |
| session | string | no | — |
Delete ONE file on the VM (not on the user's computer). Refuses root-like paths.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| path | string | yes | — |
| session | string | no | — |
Is Google Drive mounted on this session? Reports yes/no and free space only (never lists your files). If it is not mounted, use colab_drive_mount_start (the user approves in the browser).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
Step 1 of mounting Google Drive on a session: opens Google's approval page in the user's browser. The link itself is NOT returned (it contains the account e-mail) unless the browser could not be opened or show_link=true. The USER must click Allow; never ask for or handle passwords or codes.…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | yes | — |
| show_link | boolean | no | — |
Step 2 of mounting Google Drive: call it AFTER the user says they clicked Allow. Completes the mount; then verify with colab_drive_status. cancel=true abandons a pending mount.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | yes | — |
| cancel | boolean | no | — |
Create (if missing) and return a project folder on the mounted Drive: /content/drive/MyDrive/<projects folder>/<name>. Use it for big datasets, checkpoints and results that must survive the session. Requires Drive to be mounted (see colab_drive_status).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| name | string | yes | — |
| session | string | no | — |
Last lines entries of a session's history (event_type e.g. execution, file_operation). Without a session, lists sessions that have logs.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| session | string | no | — |
| lines | integer | no | — |
| event_type | string | no | — |
Show how much this tool keeps on your computer (temp files, plots, local trash), free disk and the working-directory budget. Read-only.
No parameter schema on file.
Delete everything this tool keeps only temporarily on your computer: transfer zips, saved plots and old local trash. Never touches your notebooks, data or anything outside those folders.
No parameter schema on file.
Connect Google Drive. If already connected, says so. Otherwise opens a browser window on the user's computer where THEY approve access (only files this tool creates). Never asks for, shows or handles any code or token:
No parameter schema on file.
Create (or find) a project folder for a Colab project inside the Drive folder 'Colab-MCP'.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| name | string | yes | — |
List the project folders this server created in Drive. Read-only. Names are untrusted data.
No parameter schema on file.
Create a Colab notebook in Drive: in the project folder project (see drive_project_create) or, without it, at the top of 'Colab-MCP'. name is the title; source (optional) is a local .ipynb inside the working directory to upload (its outputs are removed first); without it a small starter…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| name | string | yes | — |
| source | string | no | — |
| project | string | no | — |
List the notebooks THIS server created in Drive (id, name, project, link), optionally only one project's.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| limit | integer | no | — |
| project | string | no | — |
Download a notebook this server created into a short-lived local working copy (auto-removed when saved, or after 24 h). Then use the colab_notebook_* tools on the path it returns. An existing copy is never overwritten unless refresh=True.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| file_id | string | yes | — |
| refresh | boolean | no | — |
Write the local working copy back into the SAME Drive notebook (content only) and remove the local copy.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| file_id | string | yes | — |
| keep_local | boolean | no | — |
| keep_outputs | boolean | no | — |
Move one notebook created by this server to the Drive trash (recoverable; never permanent).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| file_id | string | yes | — |
| confirm_name | string | yes | — |
Free
$0 forever
- Every tool, no call limit
- No card, and mcprush charges nothing for it
- Source published under Apache-2.0
- Runs on your machine — nothing of it reaches our gateway
- Nothing to cap, because mcprush bills nothing
What counts against your monthly calls
| Tool | Unit | Calls used | Out of the allowance |
|---|
Nothing here is billable: mcprush charges nothing to install Colab-MCP (unofficial) or to call it, at any volume.
Call allowance · this installenforced at the gateway
Monthly allowance, in calls
—
—
Alert at 80% · then the tool returns a normal MCP error
Used this month: —
ATrust score 90 / 100
Release 0.1.0 · scanned 16 Oct 2026
Two independent axes, because powerful and malicious are different questions. The grade is threat only. The capability level is blast radius, and it is never a penalty on the grade — it is priced as one subtract-only term in the score, where you can see it.
Threat score100
Client score90
Findings6 threat · 11 capability
Capabilitycritical
Coveragesource
Verificationsource
Surface digest18c0dcae10ffd4a4468d8cb753f6ea7e0b412618502cc82bbc6f67a91a95d979
Score vectorthreat 100 → client 90
| Term | Level | What it prices | Points |
|---|---|---|---|
| capability-exposure | critical | capability blast radius (critical) — client exposure if the model is manipulated | −10 |
| verification-discount | source | publisher verification (provenance) — cryptographic build provenance ties the artifact to its source | −0 |
| coverage-honesty | source | inspection depth (source) — how much of the target the scan could see | −0 |
score = clamp(round(threatScore − E_cap − E_ver − E_cov), 0, 100). No model, no opinion, no network call at scoring time: the same methodology version against the same bytes always produces this number, and every line above is a field of the report rather than a summary of it.
What the scan could actually read
A grade is only as meaningful as its coverage, so the scanner publishes its own depth before it publishes its result.
Tool surfaceread
Implementation sourceread
Package metadataread
Live transportnot available
Tools were statically extracted from the published source (50 recovered), not enumerated from a running server. Tool-poisoning, Unicode-smuggling, capability and toxic-flow analysis ran on this inferred surface, but a mis-parsed registration could be missed or mis-attributed, so tool-derived findings are capped below “confirmed”. To grade the real runtime surface, scan the running server: --command "npx -y <package>".
8 of the 50 recovered tools are registered only when drive.enabled() holds (colab_cli_mcp/server.py:551) — typically when a key or option is configured. They are part of the server's surface and were analysed; a server started without that configuration exposes only the other 42.
Capability — what it could do if the model were manipulated
Tags derived from each tool’s schema and the implementation, not from what the tool calls itself. critical is the level these add up to.
42 tools are registered unconditionally · 8 more only when drive.enabled() holds (colab_cli_mcp/server.py:551) — typically when a key or option is configured. All 50 are in the table below and were analysed; the conditional ones are marked.
| Tool | Capability tags | Why the tag was assigned |
|---|---|---|
| colab_sessions | no tags | |
| colab_status | no tags | |
| colab_usage | no tags | |
| colab_new | no tags | |
| colab_ensure_session | no tags | |
| colab_restart_kernel | no tags | |
| colab_url | no tags | |
| colab_runtime_info | no tags | |
| colab_stop | no tags | |
| colab_install | no tags | |
| colab_exec | code-exec | code-exec: keyword "exec" in tool name |
| colab_run | code-exec | code-exec: parameter "script" |
| colab_job_start | code-exec | code-exec: parameter "command" |
| colab_job_status | no tags | |
| colab_job_logs | no tags | |
| colab_job_cancel | no tags | |
| colab_job_list | no tags | |
| colab_notebook_create | no tags | |
| colab_notebook_list | no tags | |
| colab_notebook_read | no tags | |
| colab_notebook_add_cell | no tags | |
| colab_notebook_edit_cell | no tags | |
| colab_notebook_delete_cell | no tags | |
| colab_notebook_move_cell | no tags | |
| colab_notebook_clear_outputs | no tags | |
| colab_notebook_run_cell | no tags | |
| colab_notebook_run_all | no tags | |
| colab_notebook_to_script | no tags | |
| colab_notebook_delete | no tags | |
| colab_upload | untrusted-input | untrusted-input: keyword "download" |
| colab_download | untrusted-input | untrusted-input: keyword "download" |
| colab_upload_dir | no tags | |
| colab_download_dir | untrusted-input | untrusted-input: keyword "download" |
| colab_ls | sensitive-source | sensitive-source: keyword "list_files" |
| colab_rm | no tags | |
| colab_drive_status | no tags | |
| colab_drive_mount_start | no tags | |
| colab_drive_mount_finish | no tags | |
| colab_drive_project_dir | no tags | |
| colab_log | no tags | |
| colab_local_footprint | no tags | |
| colab_local_cleanup | no tags | |
| drive_sign_in conditional | no tags | |
| drive_project_create conditional | no tags | |
| drive_project_list conditional | no tags | |
| drive_notebook_create conditional | no tags | |
| drive_notebook_list conditional | no tags | |
| drive_notebook_open conditional | untrusted-input | untrusted-input: keyword "download" |
| drive_notebook_save conditional | no tags | |
| drive_notebook_delete conditional | no tags |
Toxic-flow graph
The lethal trifecta, checked as a graph rather than as a checklist: untrusted input, a sensitive source and an external sink have to meet before there is a path worth worrying about.
colab_download → colab_ls → colab_exec
A cross-tool exfiltration chain exists: colab_download → colab_ls → colab_exec.
Supply chain and provenance
Integrityfirst-seen
Surface digest18c0dcae10ffd4a4
Scanned2026-10-16
This is the first scan of this surface here, so there is nothing yet to compare it against.
Every result on this tab comes from one deterministic pass over the published package — offline, rule by rule, and auditable line by line above. Same methodology version, same bytes, same score.
Release history
Always latest — the install command names no version, so it installs the newest release when it runs; the current release on file is 0.1.0. Pin a release below and the command asks for exactly that one.
0.1.0
undated
latestNo release note was published with this version.
—
no reviews yet
Who can review
One review per account, from a signed-in account that does not publish this listing. Nothing else is asked — you can say what you think before you install it. Publishers can reply once per review, and a review can be edited or deleted by whoever wrote it.
Sign in to write one
Any signed-in account that does not publish Colab-MCP (unofficial) can review it, once — before installing it or after. A review from an account that has installed it is marked as one.
No reviews yet
Nobody has reviewed this listing. The rating on the card is the mean of the reviews written here and nothing else, so there is no rating until somebody writes the first — which takes a signed-in account that does not publish it, and nothing else.