# io.github.namyunsu/crewdroid (npm · crewdroid)

Light Android emulators for AI agents: text screen outline, per-device network, battery, location.

- Trust score: 82/100 (high trust)
- Change this week: +16
- Registry status: active
- Liveness: live
- Owner verified: no
- Last scored: 2026-11-04

## Components

- npm · `crewdroid`: 82/100 (this document), [markdown](https://verifymcp.io/servers/namyunsu-crewdroid/crewdroid.md), [page](https://verifymcp.io/servers/namyunsu-crewdroid/crewdroid)

## Channel facts

- Registry: `npm`
- Package: `crewdroid`
- Version: `0.2.6`
- 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 2026-11-04.

- **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.
  - 31 of 121 dependencies flagged as unhealthy.
- **Provenance & Transparency**: 100/100
  - Source repository is publicly reachable at the declared URL.
  - Cryptographically verified build provenance (signed, bound to namyunsu/crewdroid).
  - Clear OSI-approved license (MIT).
  - Actively maintained (last published 0 days ago).
  - Publishes a security disclosure policy (SECURITY.md).
- **Schema Quality & AI Usability**: 60/100
  - AI-judged instruction clarity (good).
  - Context-footprint check failed: tool/resource definitions use about 3953 tokens (~152/item across 26 items; 26 tools + 0 resources), over budget; trim descriptions and params.
  - Usage-examples check failed: none of the tools include examples.
- **Stability & Change Management**: 30/100
  - Stability observed for 9 of 30 days with no destabilising changes; credit accrues until the full window elapses.
- **Tool Coverage**: 88/100
  - 100% of tools have a non-trivial description (not blank, and not just the tool's name).
  - 63% of tool parameters carry a description.
- **Tool Safety**: 100/100
  - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.
  - We read all 26 captured tool definition(s), and no name or description among them implies an irreversible operation.
  - An AI judge read all 26 captured unit(s) of tool text and found none that tries to manipulate the model reading it.
- **Capabilities**: 100/100
  - Implements a supported MCP spec version (2025-11-25); the latest is 2026-07-28.

## Install

### How do I install the io.github.namyunsu/crewdroid MCP server?

io.github.namyunsu/crewdroid runs locally as an npm package, launched with npx -y crewdroid. 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 namyunsu-crewdroid -- npx -y crewdroid
```

### Cursor

```json
{
  "mcpServers": {
    "namyunsu-crewdroid": {
      "command": "npx",
      "args": [
        "-y",
        "crewdroid"
      ]
    }
  }
}
```

### VS Code

```json
{
  "servers": {
    "namyunsu-crewdroid": {
      "command": "npx",
      "args": [
        "-y",
        "crewdroid"
      ]
    }
  }
}
```

### Codex

```bash
codex mcp add namyunsu-crewdroid -- npx -y crewdroid
```

### opencode

```json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "namyunsu-crewdroid": {
      "type": "local",
      "command": [
        "npx",
        "-y",
        "crewdroid"
      ],
      "enabled": true
    }
  }
}
```

### OpenClaw

```bash
openclaw mcp add namyunsu-crewdroid --command npx --arg -y --arg crewdroid
```

### Hermes

```yaml
mcp_servers:
  namyunsu-crewdroid:
    command: "npx"
    args: ["-y", "crewdroid"]
```

### Netclaw

```json
{
  "McpServers": {
    "namyunsu-crewdroid": {
      "Transport": "stdio",
      "Command": "npx",
      "Arguments": [
        "-y",
        "crewdroid"
      ]
    }
  }
}
```

### Vellum

```bash
assistant mcp add namyunsu-crewdroid -t stdio -c npx -a -y crewdroid
```

### Other

```json
{
  "mcpServers": {
    "namyunsu-crewdroid": {
      "command": "npx",
      "args": [
        "-y",
        "crewdroid"
      ]
    }
  }
}
```

## 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.

### 2026-11-04 (score 82, +26)

- [security improvement] Malware scan: unverified → pass
- [security improvement] Known CVEs: unverified → pass
- [functional improvement] Dependency health: unverified → 0.88

### 2026-11-03 (score 56, −14)

- [security regression] Malware scan: pass → unverified
- [security improvement] Known CVEs: unverified → pass
- [functional improvement] Dependency health: unverified → 0.88
- [functional] Package version: 0.2.2 → 0.2.6
- [functional] Package version: 0.2.2 → 0.2.4
- [functional] Package version: 0.2.2 → 0.2.3

### 2026-11-02 (score 70, −11)

- [security regression] Known CVEs: pass → unverified
- [security regression] Malware scan: pass → unverified
- [functional regression] Dependency health: 0.88 → unverified
- [functional] Package version: 0.2.0 → 0.2.2
- [functional] Package version: 0.2.0 → 0.2.1

### 2026-11-01 (score 81, +1)

No change was recorded against any check on this day. Stability & Change Management went from 17 to 20. That category is still filling its 30-day observation window: 5 days of observed history at the previous scan, 6 at this one. The score rises as the window fills, whether or not the server changes.

### 2026-10-30 (score 80, +14)

- [security improvement] Provenance: fail → pass
- [security] The attested source repository moved: namyunsu/crewdroid
- [functional regression] Schema quality: pass → fail
- [functional improvement] Tool coverage: 47% → 63%
- [functional improvement] Security disclosure: fail → pass
- [functional] Schema quality: fair → good
- [functional] Destructive annotations: 0 → pass
- [functional] Package version: 0.1.0 → 0.2.0

### 2026-10-28 (score 66, +1)

No change was recorded against any check on this day. Stability & Change Management went from 3 to 7. That category is still filling its 30-day observation window: 1 days of observed history at the previous scan, 2 at this one. The score rises as the window fills, whether or not the server changes.

### 2026-10-26 (score 65)

First indexed and scored.

## MCP tools (26)

### `devices` (~57 tokens)

Use to find the emulator-<port> serial every other tool needs, or to check what is already running. Returns [{serial, avd, api, profile, screen, cores, ramMB, booted}]; empty = call up.

### `up` (~329 tokens)

Start of the loop: boot headless emulators (1-3 min). No args = ONE device (s24ultra, newest API <= 36), the right choice for building/checking an app; count > 1 mixes Galaxy profiles and Android versions. Returns {devices:[{serial, api, profile, wmSize}]}; pass serial to every other tool.

Input parameters:

- `allowNoHelper` (boolean): true = keep the devices when the on-device helper fails (screen/tap/type then use the slower uiautomator/input)
- `api` (integer): pin the image API, e.g. 31|34|36
- `basePort` (integer)
- `cores` (integer): guest CPUs, e.g. 1|2|4|6|8 (default 2; <= host CPUs)
- `count` (integer)
- `idleMinutes` (integer): devices stop after this many minutes without a crewdroid call, and when this MCP server exits (default 60; 0 = never)
- `load` (string): used-phone load applied after boot (see the load tool)
- `profiles` (array): profile names from profiles/galaxy.json
- `ram` (integer): guest RAM MB, 896..8192 (default 896, API 31 768)
- `timeoutSec` (integer)
- `wifi` (boolean): true = each device can get its own Wi-Fi address (31 of 33 measured device boots; default: one device at a time; +9.4 % idle RAM)

### `down` (~39 tokens)

End of the loop: kill every crewdroid (crew-*) emulator. Always call it when finished. Returns {killed: pids, ports}.

### `net` (~171 tokens)

Set one emulator's network state (airplane|nodata|slow|flaky|online), which link is up (wifi|cellular|both|none), switch the active link while traffic flows (handover), or mark Wi-Fi metered. Returns the transport apps see: {default: wifi|cellular|none, connected, metered}.

Input parameters:

- `during` (integer): handover: ms with no link between old and new (default 0 = new link up before the old one goes)
- `metered` (boolean): state=metered: true = Wi-Fi metered
- `seed` (integer)
- `serial` (string, required): emulator-<port>
- `state` (string, required)
- `to` (string): handover target (default: the other link)

### `judge` (~101 tokens)

Judge server data (unique / count / noLostUpdate) and/or a proxy request log (dupSuccess / lostUpdate / lostWrite / idle) against a rules file. ok=false lists violations.

Input parameters:

- `log` (string): proxy JSONL log (proxy tool / run with proxy), e.g. with judge/proxy.rules.json
- `rules` (string)
- `source` (string): server base URL (GETs /export) or JSON file

### `hammer` (~70 tokens)

Virtual users hit a server API concurrently, then judge the result.

Input parameters:

- `jitter` (number)
- `repeat` (integer)
- `retryRate` (number)
- `rules` (string)
- `seed` (integer)
- `url` (string, required)
- `users` (integer)

### `run` (~278 tokens)

Run a multi-device scenario on any app: up, install, per-device actions (Maestro flow / intent / adb) + network schedule, drain, judge (proxy log by default), down (minutes). Give `file` (scenario JSON, docs/guide-your-app.md) OR, with no file, apk + flows + server (setupFlow, count, repeat, network optional).

Input parameters:

- `apk` (string): no-file form: your APK (package read from it)
- `basePort` (integer)
- `count` (integer): default: the file's devices.count, else 3
- `file` (string): scenario JSON, e.g. scenarios/photo-upload.json; omit to use apk/flows/server
- `flows` (array): no-file form: Maestro flows run in order every iteration
- `network` (boolean): no-file form: false = no network faults (default: all four states)
- `repeat` (integer): no-file form: iterations per device (default 5)
- `seed` (integer)
- `server` (string): no-file form: backend as the host sees it, http://host:port
- `serverPort` (integer)
- `setupFlow` (string): no-file form: Maestro flow run once per device first
- `timeoutSec` (integer)

### `apk` (~53 tokens)

Read an APK's manifest on the host (aapt2): package, version, min/target SDK, permissions, exported activities, deep-link filters.

Input parameters:

- `file` (string, required): path to the .apk on the host

### `app` (~174 tokens)

Control an app on a crewdroid emulator: perm grant|revoke|show [perm…], open <uri> (deep link / App Link), links show|force <domain>|reset (API 31+), clear, backup, restore (LocalTransport), pressure doze|standby <active|working_set|frequent|rare|restricted>|trim <level>|bg-restrict|battery-saver|reset, fill-storage <GB>|free. Setters read the state back and error on mismatch.

Input parameters:

- `args` (array)
- `onlyThisPackage` (boolean): open only: restrict the intent to pkg
- `pkg` (string, required): application id, e.g. dev.crewdroid.probe
- `serial` (string, required): emulator-<port>
- `what` (string, required)

### `install` (~101 tokens)

Use after each build: adb install -r [-g] a host .apk on one emulator or `all` crew devices (replaces the old version, keeps data). Returns ok and per device package + Success or the adb failure code (INSTALL_FAILED_*).

Input parameters:

- `apk` (string, required): host path to the .apk
- `grant` (boolean): -g: grant all runtime permissions
- `target` (string, required): emulator-<port> or all

### `launch` (~133 tokens)

Use to open the app after install, or restart it (restart:true) after a change. am start -W, then checks the process 2 s later. Returns component, totalTime ms and crashed; crashed:true comes with crash {exception, topFrame} (then call logs or triage).

Input parameters:

- `activity` (string): e.g. .MainActivity; default = launcher activity
- `extras` (array): am start extras, e.g. ["--es","check","crash"]
- `pkg` (string, required)
- `restart` (boolean)
- `serial` (string, required): emulator-<port>

### `logs` (~205 tokens)

Use to see what the app printed or why it failed: one app's logcat lines only (its pid; its last pid when not running). Default: the last 40 lines at level >= I, one text line each `HH:MM:SS.mmm L Tag: message`, equal lines in a row folded `(xN)`, stack traces cut to the top 8 app frames. all:true = today's whole output at level V (max 500 lines); since:"1m" limits to recent lines; format:"json" = {lines:[{time, level, tag, message}]}.

Input parameters:

- `all` (boolean)
- `format` (string)
- `level` (string): default I (V with all)
- `pkg` (string, required)
- `seconds` (integer): follow for N s
- `serial` (string, required): emulator-<port>
- `since` (string): 60s | 90 | 2m

### `env` (~173 tokens)

Change or read a crewdroid emulator's environment: geo <lon> <lat> [alt], battery <0-100> [ac|unplugged], rotate <0-3>, sensor acceleration x:y:z, dark on|off, font <scale>, locale <pkg> <tags> (API 33+), gsm data <home|roaming|searching|denied|off|on>, gsm signal <0-31>, sms <from> <text>, call <number> [accept|cancel], show. Every setter reads the state back and errors on mismatch.

Input parameters:

- `args` (array)
- `holdSeconds` (integer): gsm signal only: keep re-sending for this long
- `serial` (string, required): emulator-<port>
- `what` (string, required)

### `fold` (~71 tokens)

Fold (cover screen) or unfold (main screen) a fold/flip-profile emulator: the app gets a configuration change with the new size. No hinge/posture emulation on ATD.

Input parameters:

- `serial` (string, required): emulator-<port>
- `to` (string, required): cover = folded, main = unfolded

### `tap` (~224 tokens)

Use to press a button or field: by text/content-desc (text), resource-id (id), a point x,y, or index alone = the [n] line of the last screen outline (no new tree read while that window still has focus). Element match = exact first, then case-insensitive contains. 0 or several matches (text/id without index) fail and list the candidates; retry with index (0-based among the matches) or id. Returns the tapped point and the new screen outline once it settled (screen:false to skip), so no separate screen call is needed.

Input parameters:

- `format` (string)
- `id` (string): resource-id, full or the part after :id/
- `index` (integer): alone: [n] of the last screen; with text/id: 0-based among the matches
- `screen` (boolean): false = do not return the new screen
- `serial` (string, required): emulator-<port>
- `text` (string)
- `x` (integer)
- `y` (integer)

### `type` (~131 tokens)

Use to fill a text field: appends any text (Korean, emoji included) to the focused field through the on-device helper; into = text or resource-id of the field to tap first. Returns the field typed into and the new screen outline (screen:false to skip). Without the helper it falls back to `input text`, where non-ASCII fails with the reason.

Input parameters:

- `format` (string)
- `into` (string)
- `screen` (boolean): false = do not return the new screen
- `serial` (string, required): emulator-<port>
- `text` (string, required)

### `swipe` (~145 tokens)

Use to scroll or page: swipe up|down|left|right (finger direction, middle 50 % of the screen; up scrolls content down) `times` times, or a line x1,y1 -> x2,y2 in ms. Returns the from/to points and the new screen outline (screen:false to skip).

Input parameters:

- `direction` (string)
- `format` (string)
- `line` (array): [x1,y1,x2,y2]
- `ms` (integer)
- `screen` (boolean): false = do not return the new screen
- `serial` (string, required): emulator-<port>
- `times` (integer)

### `key` (~86 tokens)

Use for navigation and hardware keys: back|home|enter|delete|tab|menu or any KEYCODE_*. Returns the focused window and the new screen outline (screen:false to skip).

Input parameters:

- `format` (string)
- `key` (string, required)
- `screen` (boolean): false = do not return the new screen
- `serial` (string, required): emulator-<port>

### `layout` (~349 tokens)

Use to find broken screens of any installed app across Galaxy screens x font sizes on ONE device (default 5 profiles x fonts 1.0/1.3/2.0 = 15 combos, ~2-4 min). Per combo the app is restarted and the launch activity (or the screen a Maestro flow reaches) is judged from the accessibility tree: wrap (button label on 2+ lines, word broken), clip (ellipsized/cut text), offscreen (no scroll can reach it), overlap (> 20 %), target (< 48 dp). Returns {ok, combos:[{profile, font, ok, findings:[{kind, text|id, bounds, detail}], png?}], summary}; png = screenshot with the finding boxes drawn. The device's screen size, density and font are restored afterwards.

Input parameters:

- `activity` (string): e.g. .MainActivity; default = launcher activity
- `extras` (array): am start extras, e.g. ["--es","check","layout-bad"]
- `flow` (string): Maestro flow (host path) run after each start; the screen it ends on is judged
- `fonts` (array): font scales, default [1.0,1.3,2.0]
- `out` (string): directory for the PNGs (default ./crewdroid-layout-<pkg>-<time>)
- `pkg` (string, required): installed package
- `profiles` (array): profile names, default s24ultra,fold5-cover,fold5-main,flip6-cover,trifold-main
- `serial` (string, required): emulator-<port>
- `timeoutSec` (integer)

### `load` (~152 tokens)

Make one emulator behave like a used phone: off|low|mid|high = installed filler apps (30/80/150), background processes (5/15/30), memory ballast down to 50/30/15 % of the off-level free RAM, CPU workers at 10/25/50 % per core, alarm/job wakeups. off removes it all (keepApps keeps the fillers installed). Returns the measured guest state (MemAvailable %, load average, CPU busy %, processes, packages).

Input parameters:

- `keepApps` (boolean): off only: keep the filler packages installed
- `level` (string, required)
- `serial` (string, required): emulator-<port>
- `timeoutSec` (integer)

### `perf` (~118 tokens)

Cold start time (force-stop, 2 s, am start -W), frames/jank %/p50-p99 (gfxinfo) and PSS java/native (meminfo) of one app. ATD has no GPU: compare frame numbers relatively.

Input parameters:

- `activity` (string): e.g. .MainActivity; default = launcher activity
- `extras` (array): am start extras, e.g. ["--es","check","jank"]
- `pkg` (string, required)
- `serial` (string, required): emulator-<port>

### `proxy` (~175 tokens)

Recording HTTP proxy between the emulators and your server (no app or server change): action start points every device's system proxy at it (10.0.2.2:<port+k>), stop clears it, log returns the request log. Judge the log with the judge tool's `log` input (dupSuccess / lostUpdate / lostWrite / idle).

Input parameters:

- `action` (string, required)
- `devices` (array): default: every running crew emulator
- `log` (string): JSONL path
- `loseAcks` (number): fault injection: share of writes whose answer is dropped after the server committed them
- `port` (integer)
- `target` (string): start: server base URL as seen from the host, e.g. http://127.0.0.1:5599

### `screen` (~237 tokens)

Use to READ the screen before tap/type (tree, not pixels). Default: a short outline, first line = focused window and size, then one line per visible element `[n] role "text" #id @x,y flags` (role btn/edit/text/img/list/check; a card's texts sit on its one line; `… N more below (swipe up)` = off-screen). tap {index:n} alone taps [n] of this outline. format:"json" = full tree {focus, nodes[{cls,text,desc,id,bounds,clickable,depth,...}], via}. While a Maestro flow runs the tree is empty and a line says why. For how it looks, use shot. On API 31 a black screen after a relaunch (emulator bug, ADR 0005) is recovered first and reported (blackRelaunch), not an app bug.

Input parameters:

- `bounds` (boolean): add [l,t,r,b] to each line
- `format` (string)
- `png` (string): host path for a screenshot
- `serial` (string, required): emulator-<port>

### `setup` (~161 tokens)

Use first on a new machine, or when up fails for a missing SDK/image: checks/installs the Android SDK pieces crewdroid needs (JDK 17 for sdkmanager, command-line tools, platform-tools, emulator, google_atd images). Dry run by default: returns steps [{step, status: present|missing, downloadBytes}]; yes:true installs and accepts the SDK licences.

Input parameters:

- `images` (array): Android API levels of google_atd images, default ["34"]
- `only` (array): limit to these steps
- `sdk` (string): SDK directory (default ANDROID_SDK_ROOT / ANDROID_HOME, else ~/.crewdroid/android-sdk)
- `yes` (boolean): install for real and accept the SDK licences

### `shot` (~118 tokens)

Use to SEE the screen (pixels: layout, colours, clipping). Returns an image item (PNG, also saved to file or ./crewdroid-<serial>-<time>.png) and a text line with path and size. To find a button or text to tap, use screen instead. On API 31 a fully black screen after an app relaunch (emulator bug, ADR 0005) is recovered first.

Input parameters:

- `file` (string): host path for the PNG
- `serial` (string, required): emulator-<port>

### `triage` (~102 tokens)

Why did the app die? Crashes (logcat crash buffer), ANRs (ActivityManager log, exit-info, /data/anr traces), exit reasons, the app's last 200 log lines, optional heap dump.

Input parameters:

- `app` (string): package name, e.g. dev.crewdroid.probe
- `heap` (string): host path for an .hprof (needs app running)
- `serial` (string, required): emulator-<port>

## Diagnostics

Captured diagnostic sections: Provenance, Dependencies. The full working is on the page: https://verifymcp.io/servers/namyunsu-crewdroid/crewdroid#diagnostics

## Score history

- 2026-11-04: 82
- 2026-11-03: 56
- 2026-11-02: 70
- 2026-11-01: 81
- 2026-10-31: 80
- 2026-10-30: 80
- 2026-10-29: 66
- 2026-10-28: 66
- 2026-10-27: 65
- 2026-10-26: 65

## Common questions

### What is the io.github.namyunsu/crewdroid MCP server?

io.github.namyunsu/crewdroid is an MCP server listed in the public MCP registry as io.github.namyunsu/crewdroid. Light Android emulators for AI agents: text screen outline, per-device network, battery, location. This page covers its npm package (crewdroid).

### Is the io.github.namyunsu/crewdroid MCP server safe to use?

io.github.namyunsu/crewdroid scores 82 out of 100 on VerifyMCP. We found no known CVEs affecting it as of 4 November 2026. It declares no install or post-install scripts. Its build provenance is signed and verified. 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 io.github.namyunsu/crewdroid MCP server expose?

io.github.namyunsu/crewdroid exposes 26 tools: devices, up, down, net, judge, and 21 more. Their descriptions and schemas cost roughly 3,953 tokens of context every time the server is loaded.

### Is the io.github.namyunsu/crewdroid MCP server still maintained?

io.github.namyunsu/crewdroid is still listed as active in the MCP registry. We last reached this channel on 4 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 io.github.namyunsu/crewdroid MCP server under?

io.github.namyunsu/crewdroid declares the MIT licence, which is OSI-approved. That covers the source only, and says nothing about the cost of any service it calls.

## Links

- npm package: https://www.npmjs.com/package/crewdroid
- Socket report: https://socket.dev/npm/package/crewdroid
- Repository: https://github.com/namyunsu/crewdroid
- Changelog RSS feed: https://verifymcp.io/servers/namyunsu-crewdroid/crewdroid.xml
- Changelog JSON feed: https://verifymcp.io/servers/namyunsu-crewdroid/crewdroid.json
- HTML version of this page: https://verifymcp.io/servers/namyunsu-crewdroid/crewdroid