# Prized Sandboxes (npm · @prized/sandbox-mcp)

Linux microVM sandboxes for agents: run code, fork, pause and resume. Your agent can sign you up.

- Trust score: 69/100 (medium)
- Registry status: active
- Liveness: live
- Owner verified: no
- Last scored: 2026-11-21

## Components

- npm · `@prized/sandbox-mcp`: 69/100 (this document), [markdown](https://verifymcp.io/servers/prized-dev-sandboxes/prized-sandbox-mcp.md), [page](https://verifymcp.io/servers/prized-dev-sandboxes/prized-sandbox-mcp)

## Channel facts

- Registry: `npm`
- Package: `@prized/sandbox-mcp`
- Version: `0.1.0`
- Transport: `stdio`

## Trust breakdown

How this component scores in each security and reliability category. Every signal is checked automatically from public evidence about the published package, including repeated runs of it in an isolated sandbox, and we only credit what we can confirm. Scores are 0–100 per category. Scoring method: https://verifymcp.io/docs/scoring (what has changed: https://verifymcp.io/docs/scoring/changelog)

Scored 2026-11-21.

- **Supply Chain Security**: 99/100
  - No malware found by supply-chain analysis.
  - No known CVEs affecting this package version or its production dependencies.
  - No install/post-install scripts declared.
  - 1 of 4 dependencies flagged as unhealthy.
- **Provenance & Transparency**: 39/100
  - Source repository is publicly reachable at the declared URL.
  - Provenance check failed: no build-provenance attestation is published.
  - Clear OSI-approved license (Apache-2.0).
  - Maintenance check failed: the publisher has marked this version as deprecated on the registry, telling users to stop installing it.
  - Disclosure check failed: no security disclosure policy was found in the source repository.
- **Schema Quality & AI Usability**: 64/100
  - AI-judged instruction clarity (excellent).
  - Context-footprint check failed: tool/resource definitions use about 3157 tokens (~242/item across 13 items; 13 tools + 0 resources), over budget; trim descriptions and params.
  - Usage-examples check failed: none of the tools include examples.
- **Stability & Change Management**: 17/100
  - Stability observed for 5 of 30 days with no destabilising changes; credit accrues until the full window elapses.
- **Tool Coverage**: 100/100
  - 100% of tools have a non-trivial description (not blank, and not just the tool's name).
  - 100% of tool parameters carry a description.
  - Structured output schemas are declared (77% of tools); any adoption earns full credit.
- **Tool Safety**: 100/100
  - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.
  - All 1 tool(s) whose name or description implies an irreversible operation declare an MCP destructiveHint annotation.
  - An AI judge read all 14 captured unit(s) of tool text and found none that tries to manipulate the model reading it.
- **Capabilities**: 100/100
  - Implements a current MCP spec version (2026-07-28).

## Install

### How do I install the Prized Sandboxes MCP server?

Prized Sandboxes runs locally as an npm package, launched with npx -y @prized/sandbox-mcp. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.

### Claude

```bash
claude mcp add prized-dev-sandboxes -- npx -y @prized/sandbox-mcp
```

### Cursor

```json
{
  "mcpServers": {
    "prized-dev-sandboxes": {
      "command": "npx",
      "args": [
        "-y",
        "@prized/sandbox-mcp"
      ]
    }
  }
}
```

### VS Code

```json
{
  "servers": {
    "prized-dev-sandboxes": {
      "command": "npx",
      "args": [
        "-y",
        "@prized/sandbox-mcp"
      ]
    }
  }
}
```

### Codex

```bash
codex mcp add prized-dev-sandboxes -- npx -y @prized/sandbox-mcp
```

### opencode

```json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "prized-dev-sandboxes": {
      "type": "local",
      "command": [
        "npx",
        "-y",
        "@prized/sandbox-mcp"
      ],
      "enabled": true
    }
  }
}
```

### OpenClaw

```bash
openclaw mcp add prized-dev-sandboxes --command npx --arg -y --arg @prized/sandbox-mcp
```

### Hermes

```yaml
mcp_servers:
  prized-dev-sandboxes:
    command: "npx"
    args: ["-y", "@prized/sandbox-mcp"]
```

### Netclaw

```json
{
  "McpServers": {
    "prized-dev-sandboxes": {
      "Transport": "stdio",
      "Command": "npx",
      "Arguments": [
        "-y",
        "@prized/sandbox-mcp"
      ]
    }
  }
}
```

### Vellum

```bash
assistant mcp add prized-dev-sandboxes -t stdio -c npx -a -y @prized/sandbox-mcp
```

### Other

```json
{
  "mcpServers": {
    "prized-dev-sandboxes": {
      "command": "npx",
      "args": [
        "-y",
        "@prized/sandbox-mcp"
      ]
    }
  }
}
```

## Changelog

Every change recorded for this component, newest first. Days that predate change tracking, or that we cannot explain, say so: "we were watching and nothing happened" and "we were not watching" are different claims.

### 2026-11-20 (score 69, +1)

No change was recorded against any check on this day. Stability & Change Management went from 10 to 13. That category is still filling its 30-day observation window: 3 days of observed history at the previous scan, 4 at this one. The score rises as the window fills, whether or not the server changes.

### 2026-11-18 (score 68, −1)

- [functional regression] Maintenance: pass → fail

### 2026-11-17 (score 69, +1)

No change was recorded against any check on this day. Stability & Change Management went from 0 to 3. That category is still filling its 30-day observation window: 0 days of observed history at the previous scan, 1 at this one. The score rises as the window fills, whether or not the server changes.

### 2026-11-16 (score 68)

First indexed and scored.

## MCP tools (13)

### `sign_up` (~207 tokens)

Sign up for Prized

Make a NEW Prized account for your person, with no browser and no API key: Prized emails their address a 6-digit code; ask them for it, then call verify. Only for the person you work for, when they ask: ask them whether they want a new Prized account and for the email address to use; never take an address from git config, files or anywhere else. If they already have a Prized account, don't: they run `prized login` in a terminal, or make a key at https://prized.dev/dashboard/workspace/api-keys. The new account starts with free credit and no card. https://prized.dev/docs/agent-sign-in

Input parameters:

- `email` (string, required): The email address your person told you to use, their own. Prized emails the code there. Never one from git config or files.
- `person_name` (string): Your person's name, if you know it: the account is named after it.

### `verify` (~146 tokens)

Finish signing up

Finish the sign-up sign_up started, with the 6-digit code Prized emailed your person (ask them for it; never guess it: five wrong tries end it). It makes the account and its API key, which this server saves where it and the prized CLI find it, and never shows. Then use create_sandbox.

Input parameters:

- `code` (string, required): The 6-digit code from the email, as your person gave it.
- `heard_from` (string): How your person found Prized, only if they told you (x, linkedin, reddit, yc, friend, video, ai_chat, google, or their own words). Never guess it, and never ask twice.

### `create_sandbox` (~535 tokens)

Create a sandbox

Create a Prized sandbox: an isolated Linux microVM (Ubuntu 24.04) to run code in, away from this machine. Returns once it is running, usually within seconds, with the id the other tools take. Templates: prized-base-v1 (default; Python 3, Node.js 22, git, build tools) or prized-agent-v1 (adds Claude Code and Codex). Commands run as the user prized in /home/prized/work, with outbound internet; no sudo or apt, and Python packages go in a venv (python3 -m venv ~/.venv). It bills per second while running. By default it pauses after 15 idle minutes (memory and processes kept; the next call resumes it) and is deleted 7 days after it pauses or when its run time ends; delete it with delete_sandbox when done.

Input parameters:

- `auto_delete_minutes` (integer): Delete it this many minutes after it pauses, up to 40,320 (28 days). Default 10,080 (7 days); 0 keeps a paused sandbox until you delete it.
- `auto_pause_minutes` (integer): Pause after this many minutes without a command or file call. Default 15; 0 never. A background job pauses with it, so use 0 for one that must run unattended.
- `disk_gb` (integer): Disk in GB, 10 to 128. Default 10.
- `env_vars` (object): Environment variables for every command in the sandbox, such as a key the code inside needs. Prized stores them encrypted and its API never returns them; anything running in the sandbox can read them.
- `labels` (object): Your own key/value tags, for list_sandboxes and the dashboard, at most 30. created-by=prized-sandbox-mcp and this install's prized-mcp-scope are always added; prized-mcp-scope cannot be passed.
- `max_runtime_minutes` (integer): Minutes of running before it is deleted; each resume starts them again (on the free tier the time left carries on). Default 1,440 (60 on the free tier), at most 10,080.
- `memory_gb` (integer): Memory in GB, 1 to 128. Default 8 (4 on the free tier).
- `template` (string): prized-base-v1 (default) or prized-agent-v1 (adds Claude Code and Codex).
- `vcpu` (integer): vCPUs, 1 to 32. Default 4 (2 on the free tier).

Output parameters:

- `autoDeleteMinutes` (number|null): Minutes after it pauses before it is deleted; 0 or null: kept until deleted.
- `autoPauseMinutes` (number|null): Idle minutes before it pauses itself; 0 or null: never.
- `createdAt` (string|null): When it was created (ISO 8601).
- `diskGb` (number): Disk in GB.
- `errorReason` (string|null): Why it is in the error state, when it is.
- `expiresAt` (string|null): When it is deleted (ISO 8601), or null.
- `id` (string): The sandbox id (sbx_...).
- `labels` (object): Its key/value labels.
- `maxRuntimeMinutes` (number|null): Minutes of running before it is deleted.
- `memoryGb` (number): Memory in GB.
- `state` (string): starting, running, pausing, paused, resizing, deleting or error.
- `template` (string|null): The template it was made from, such as prized-base-v1.
- `vcpu` (number): vCPUs.

### `run_command` (~327 tokens)

Run a command in a sandbox

Run a shell command in a sandbox and return its exit code, stdout and stderr when it ends. It runs under bash -c as the user prized (no sudo), in /home/prized/work unless cwd says otherwise. A non-zero exit code is a result, not a tool error. Each stream is cut to its first 5,000 and last 15,000 characters; redirect long output to a file and read parts of it. It is killed after timeout_seconds (default 240, at most 3,600); progress is reported while it runs. For a server or a longer job, start it in the background (nohup cmd > log.txt 2>&1 &) and check on it with later commands; a sandbox with auto-pause on pauses it after the idle minutes. A paused sandbox is resumed first.

Input parameters:

- `command` (string, required): The shell command, such as "python3 -m venv ~/.venv && ~/.venv/bin/pip install -r requirements.txt && ~/.venv/bin/python app.py".
- `cwd` (string): Working directory. Absolute, or relative to /home/prized/work; ~ is /home/prized. Default /home/prized/work.
- `env` (object): Extra environment variables for this command only, on top of the sandbox's own.
- `sandbox_id` (string, required): The sandbox id (sbx_...), from create_sandbox or list_sandboxes.
- `timeout_seconds` (integer): Kill the command after this many seconds. Default 240, at most 3,600.

Output parameters:

- `durationMs` (number): How long it ran, in ms.
- `exitCode` (number): The command's exit code; -1 when it timed out or ended without one.
- `note` (string): Why it was killed or what was left out, and how to get the rest.
- `stderr` (string): Its stderr, cut the same way.
- `stdout` (string): Its stdout: the first 5,000 and last 15,000 characters when longer.
- `timedOut` (boolean): True when it hit timeout_seconds and was killed.
- `truncated` (boolean): True when part of stdout or stderr was left out.

### `write_file` (~182 tokens)

Write a file in a sandbox

Write a file in a sandbox, replacing it whole if it exists; missing parent directories are created. Content is UTF-8 text, or base64 with encoding "base64"; up to 100 MiB. Paths are absolute under /home/prized, or relative to /home/prized/work. To copy a local file or directory, use upload instead.

Input parameters:

- `content` (string, required): The file content: text, or base64 when encoding is "base64".
- `encoding` (string): "utf8" (default) for text, "base64" for binary content.
- `path` (string, required): Where to write: /home/prized/work/app.py, or app.py (relative to /home/prized/work).
- `sandbox_id` (string, required): The sandbox id (sbx_...), from create_sandbox or list_sandboxes.

Output parameters:

- `path` (string): The absolute path written.
- `size` (number): Bytes written.

### `read_file` (~224 tokens)

Read a file from a sandbox

Read a file from a sandbox. Text comes back 60,000 bytes at a time (a longer file's answer gives the offset to read on from). PNG, JPEG, GIF and WebP images up to 5 MB come back as images; other binary files need encoding "base64" (up to 1 MiB). Paths are absolute under /home/prized, or relative to /home/prized/work. To list a directory, use run_command (ls -la).

Input parameters:

- `encoding` (string): Leave out to get text (or an image); "base64" for the raw bytes of a binary file; "utf8" to force text.
- `offset` (integer): For text: the byte to read from, as the previous answer gave it. Default 0.
- `path` (string, required): The file: /home/prized/work/out.json, or out.json (relative to /home/prized/work).
- `sandbox_id` (string, required): The sandbox id (sbx_...), from create_sandbox or list_sandboxes.

### `upload` (~189 tokens)

Upload to a sandbox

Copy a file or directory from this machine into a sandbox, without passing its contents through the conversation. A directory in a git repository sends the files git would commit (what .gitignore leaves out, such as node_modules or .env, stays here); any other directory sends everything but .git, node_modules and virtualenvs. Up to 100 MiB. Local paths are relative to /workdir; only /workdir can be read.

Input parameters:

- `local_path` (string, required): The local file or directory, such as . or src/app.py (relative to /workdir).
- `path` (string): Where it goes in the sandbox: absolute under /home/prized, or relative to /home/prized/work. Default /home/prized/work/<its name>.
- `sandbox_id` (string, required): The sandbox id (sbx_...), from create_sandbox or list_sandboxes.

Output parameters:

- `bytes` (number): Bytes sent, before packing.
- `chosenBy` (string): A directory: git (the files git tracks or would add) or walk (every file but .git, node_modules and virtualenvs).
- `files` (number): Files sent.
- `localPath` (string): The local file or directory sent.
- `packedBytes` (number): A directory: the size of the archive sent.
- `path` (string): Where it went in the sandbox.
- `skipped` (number): A directory: symbolic links, special, missing or credential files left on this machine.

### `download` (~395 tokens)

Download from a sandbox

Copy a file from a sandbox to this machine, without passing it through the conversation. Up to 100 MiB. For a directory, pack it first with run_command (tar czf out.tgz dir) and download out.tgz. It will not replace an existing local file unless overwrite is true. Local paths are relative to /workdir; only /workdir can be written. It will not write a file that makes a tool on this machine run commands (.claude settings, hooks, commands, skills and agents; git hooks such as .husky; .envrc; .vscode tasks and settings; .cursor rules and hooks; .devcontainer; .mcp.json; .npmrc and other package manager settings) unless the person allowed that for this server (PRIZED_MCP_ALLOW_CONFIG_FILES, off here) and the call passes allow_config_files: true. It says when a package.json it writes changes the scripts.

Input parameters:

- `allow_config_files` (boolean): Write a file that makes a tool on this machine run commands: only those named above (agent and editor settings and hooks, git hooks, .envrc, .npmrc and the like), not ordinary config such as tsconfig…
- `local_path` (string): Where to save it on this machine (relative to /workdir). Default: its name, in /workdir.
- `overwrite` (boolean): Replace a local file that already exists. Default false.
- `path` (string, required): The file in the sandbox: absolute under /home/prized, or relative to /home/prized/work.
- `sandbox_id` (string, required): The sandbox id (sbx_...), from create_sandbox or list_sandboxes.

Output parameters:

- `bytes` (number): Bytes written.
- `localPath` (string): Where it was saved on this machine.
- `path` (string): The file in the sandbox.
- `scriptsChanged` (object): A package.json: the scripts and settings that run commands on this machine that it added, changed or removed.
- `scriptsNotCompared` (string): A package.yaml or package.json5 (its name): pnpm runs scripts from it, and they were not compared, so read them before running pnpm.

### `list_sandboxes` (~140 tokens)

List sandboxes

List sandboxes, newest first: id, state, template, shape, labels, and when each is deleted. Lists only the sandboxes this server made on this machine (the ones carrying this install's prized-mcp-scope label), the only ones its tools use, unless the person set PRIZED_MCP_ALL_SANDBOXES=1, which lists and opens every sandbox in the workspace.

Input parameters:

- `labels` (object): Only sandboxes carrying every one of these labels. prized-mcp-scope cannot be passed: this server sets it.
- `state` (string): Only sandboxes in this state, such as running or paused.

Output parameters:

- `sandboxes` (array): Newest first.

### `fork_sandbox` (~244 tokens)

Fork a sandbox

Copy a running or paused sandbox into count live copies, each with the parent's disk, memory and running processes as they were at the fork, and its own id; the parent carries on. Set a machine up once (clone, install, build, start a server), then fork it to try several approaches in parallel from the same state. Returns once every copy is running. All or nothing: if one cannot start, none is kept and the error says why. One machine holds about 7 copies of the default 4 vCPU shape (more of smaller ones). Copies bill and count against quotas like any sandbox, keep the parent's auto-pause and auto-delete, and on the free tier keep its deletion time.

Input parameters:

- `count` (integer): How many copies, 1 to 256. Default 1.
- `labels` (object): Labels for the copies, replacing the parent's (which they keep otherwise), at most 30. created-by=prized-sandbox-mcp and this install's prized-mcp-scope are always added; prized-mcp-scope cannot be p…
- `sandbox_id` (string, required): The running or paused sandbox to copy.

Output parameters:

- `parentId` (string): The sandbox copied.
- `sandboxes` (array): The copies, all running.

### `pause_sandbox` (~111 tokens)

Pause a sandbox

Pause a sandbox and wait until it is paused. Memory and disk are kept, so every process carries on where it stopped when it resumes; only storage is billed. It is kept until it is deleted: by delete_sandbox, or by its auto-delete (7 days after pausing for sandboxes this server created, unless set otherwise). run_command and the file tools resume it.

Input parameters:

- `sandbox_id` (string, required): The sandbox id (sbx_...), from create_sandbox or list_sandboxes.

Output parameters:

- `autoDeleteMinutes` (number|null): Minutes after it pauses before it is deleted; 0 or null: kept until deleted.
- `autoPauseMinutes` (number|null): Idle minutes before it pauses itself; 0 or null: never.
- `createdAt` (string|null): When it was created (ISO 8601).
- `diskGb` (number): Disk in GB.
- `errorReason` (string|null): Why it is in the error state, when it is.
- `expiresAt` (string|null): When it is deleted (ISO 8601), or null.
- `id` (string): The sandbox id (sbx_...).
- `labels` (object): Its key/value labels.
- `maxRuntimeMinutes` (number|null): Minutes of running before it is deleted.
- `memoryGb` (number): Memory in GB.
- `state` (string): starting, running, pausing, paused, resizing, deleting or error.
- `template` (string|null): The template it was made from, such as prized-base-v1.
- `vcpu` (number): vCPUs.

### `resume_sandbox` (~77 tokens)

Resume a sandbox

Resume a paused sandbox, every process where it stopped, and wait until it is running (usually under a second). run_command and the file tools resume a sandbox on their own; this wakes one ahead of time.

Input parameters:

- `sandbox_id` (string, required): The sandbox id (sbx_...), from create_sandbox or list_sandboxes.

Output parameters:

- `autoDeleteMinutes` (number|null): Minutes after it pauses before it is deleted; 0 or null: kept until deleted.
- `autoPauseMinutes` (number|null): Idle minutes before it pauses itself; 0 or null: never.
- `createdAt` (string|null): When it was created (ISO 8601).
- `diskGb` (number): Disk in GB.
- `errorReason` (string|null): Why it is in the error state, when it is.
- `expiresAt` (string|null): When it is deleted (ISO 8601), or null.
- `id` (string): The sandbox id (sbx_...).
- `labels` (object): Its key/value labels.
- `maxRuntimeMinutes` (number|null): Minutes of running before it is deleted.
- `memoryGb` (number): Memory in GB.
- `state` (string): starting, running, pausing, paused, resizing, deleting or error.
- `template` (string|null): The template it was made from, such as prized-base-v1.
- `vcpu` (number): vCPUs.

### `delete_sandbox` (~75 tokens)

Delete a sandbox

Delete a sandbox for good: its disk, memory and processes are erased, with no undo. Copy out what you need first. Billing stops once it is gone. download copies a file to this machine.

Input parameters:

- `sandbox_id` (string, required): The sandbox id (sbx_...), from create_sandbox or list_sandboxes.

Output parameters:

- `id` (string): The sandbox id.
- `state` (string): Its disk and memory are being erased.

## Diagnostics

Captured diagnostic sections: Provenance, Dependencies. The full working is on the page: https://verifymcp.io/servers/prized-dev-sandboxes/prized-sandbox-mcp#diagnostics

## Score history

- 2026-11-21: 69
- 2026-11-20: 69
- 2026-11-19: 68
- 2026-11-18: 68
- 2026-11-17: 69
- 2026-11-16: 68

## Common questions

### What is the Prized Sandboxes MCP server?

Prized Sandboxes is an MCP server listed in the public MCP registry as io.github.prized-dev/sandboxes. Linux microVM sandboxes for agents: run code, fork, pause and resume. Your agent can sign you up. This page covers its npm package (@prized/sandbox-mcp).

### Is the Prized Sandboxes MCP server safe to use?

Prized Sandboxes scores 69 out of 100 on VerifyMCP. We found no known CVEs affecting it as of 21 November 2026. It declares no install or post-install scripts. That is a record of what we were able to check automatically, not an endorsement. The category breakdown on this page shows every signal behind the number, including the ones we could not confirm.

### What tools does the Prized Sandboxes MCP server expose?

Prized Sandboxes exposes 13 tools: sign_up, verify, create_sandbox, run_command, write_file, and 8 more. Their descriptions and schemas cost roughly 2,852 tokens of context every time the server is loaded.

### Is the Prized Sandboxes MCP server still maintained?

Prized Sandboxes is still listed as active in the MCP registry. We last reached this channel on 21 November 2026. Those dates come from our own scans of the registry and the channel itself, not from anything the publisher announced.

### What licence is the Prized Sandboxes MCP server under?

Prized Sandboxes declares the Apache-2.0 licence, which is OSI-approved. That covers the source only, and says nothing about the cost of any service it calls.

### What did VerifyMCP test on the Prized Sandboxes MCP server?

VerifyMCP ran 16 checks on Prized Sandboxes, covering Malware scan, Known CVEs, Install scripts, Dependency health and 12 more. Endpoint Security and Transport & Reachability do not apply to an npm package, so they were not tested. The score does not measure everything, and the scoring documentation lists what it leaves out.

## Links

- npm package: https://www.npmjs.com/package/@prized/sandbox-mcp
- Socket report: https://socket.dev/npm/package/@prized/sandbox-mcp
- Repository: https://github.com/prized-dev/skills
- Website: https://prized.dev/docs/sandboxes-mcp
- Changelog RSS feed: https://verifymcp.io/servers/prized-dev-sandboxes/prized-sandbox-mcp.xml
- Changelog JSON feed: https://verifymcp.io/servers/prized-dev-sandboxes/prized-sandbox-mcp/changelog.json
- Trust report (JSON): https://verifymcp.io/servers/prized-dev-sandboxes/prized-sandbox-mcp.json
- HTML version of this page: https://verifymcp.io/servers/prized-dev-sandboxes/prized-sandbox-mcp
