
Iotaudit
io.github.emirmrkaljevic由 GitHub 帳號 emirmrkaljevic 發布v0.1.18更新於 Nov 7, 2026
Audit your own smart-home devices: destinations, encryption, open ports, A-F grade.
概覽
在本機助理中稽核自家的智慧家庭裝置:連線目的地、加密狀況、開放連接埠、UPnP 連接埠轉送與 A-F 評級。
- 可以做什麼
- 依裝置分析已儲存的 pcap 或 PCAPdroid 擷取檔,可選擇輸出 PDF
- 比對兩份已儲存的 JSON 稽核結果,列出新增的目的地、國家、連接埠、評級變差或未加密流量
- 提供每台裝置的發現:目標網域、IP、國家、加密狀態、流量型態指標與離線已知問題提示
- 在報告中呈現開放連接埠、UPnP 連接埠轉送,以及閒置上傳或影片路徑的觀察結果
- 適用情境
- 適合讓助理解讀你已蒐集的智慧家庭或物聯網網路擷取檔,或比較不同時間的稽核結果並指出變化,且不希望資料上傳到雲端服務。
- 執行需求
- 以本機程序執行,從 PyPI 安裝(uvx 或含 mcp 附加項的 pip),以 iotaudit-mcp 啟動。分析既有擷取檔不需 root;即時擷取與掃描需要 sudo 以及 nmap 和 tcpdump。未宣告任何帳號、API 金鑰或環境變數。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Iotaudit,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
iotaudit
[CI] [License: MIT] [Python 3.10+] [PyPI]
[Support my open-source work: get a SkinToScan pack]
If you find this useful, you can support me by getting a SkinToScan pack, my other project. It's a one-time purchase that directly funds my open-source work.
[Buy Single Picture Check, €4.49] [Buy 5 Picture Pack, €19.99 (popular)] [Buy 15 Picture Pack, €54.99]
Quick links: sample report · what common devices send · Home Assistant · MCP server
[iotaudit auto demo (sample data)]
Is your smart pet feeder, camera or plug safe? iotaudit watches what your own IoT devices
do on the network and gives each one an A-F grade, a one-line safety verdict, and a one-page HTML report. Everything
runs locally: no account, no telemetry, no cloud.
It answers the questions you actually have about a cheap smart device:
- What is the short version? A one-line verdict next to the grade. Open ports, public DNS, TLS or plaintext, and destination countries are observed facts. Video, audio, and other traffic patterns stay indicators with a confidence.
- Who does it talk to? Domains (DNS + TLS SNI + HTTP Host), IPs, country, organisation. The report lists the destination countries for each device (observed fact).
- What does the traffic look like? A separate per-device indicator for video, audio, telemetry, or unknown, each with a low, medium, or high confidence. It is a pattern match, not proof of what was sent.
- Is it encrypted? Plaintext HTTP, unencrypted TCP, old TLS, or "custom" high-entropy protocols.
- What ports are open? Optional
nmapscan with risk notes (telnet, RTSP, UPnP, ADB, …). - Any known issue for this banner? A short offline hint when a version string matches a curated CVE. It does not change the grade unless the version match is exact and the entry is critical.
- Is it reachable from the internet? Lists the port forwards your router opened via UPnP and which device they point to (read-only, never changes the router).
- Does it phone home while idle? Idle-upload check and a traffic timeline with bursts.
- Where did the video go? LAN peer (phone P2P) vs cloud vs STUN/TURN, and an honest note when a stream was set up but never actually passed through your computer.
- What changed since the last audit?
iotaudit diffcompares two saved JSON results and exits 1 when a new destination, country, open port, worse grade, unencrypted flow, UPnP forward, or device appeared.
Status: 0.1.18. CSV export via
iotaudit export --format csv; Home Assistant summary via--ha-json.iotaudit autodiscovers devices, runs the safe checks, captures passively, and writes the report in one command. Hotspot + ARP capture, pcap and PCAPdroid analysis, IPv6/NDP, device-type identification, UPnP port-forward audit, an opt-in default-password check, grading with a one-line safety verdict, HTML and PDF reports, a diff between two audits, offline known-issue hints, a local web UI and the MCP server work. A full vulnerability feed is still on the roadmap.
Sample report
A report for a Tuya-associated pet feeder (anonymized). The one-line verdict sits beside the grade: telnet open on the LAN, public DNS, TLS cloud traffic to the US, unverified custom encryption, and a P2P video setup whose stream was not observed on this run. The longer summary is still underneath. It also shows an odd special-use "LAN peer" that is not a real device:
Open docs/sample-report.html in a browser for the full version.
Demo
A short walk-through of a capture, the UPnP port-forward check and the opt-in default-password check (animated SVG, no audio, no tracking):
Install
Capturing needs root (sudo). The optional port scan needs nmap.
Quick start
One command does the audit. It asks you to confirm this is your network, picks the active interface, finds devices, identifies them, scans ports, lists UPnP forwards (read-only), captures passively for 60 seconds, and writes the report.
Each device is printed on one line, then the report path. Files land in ~/iotaudit/YYYY-MM-DD/ as HTML and JSON. A PDF is written next to them when iotaudit[pdf] is installed (pip install "iotaudit[pdf]"). If that folder already has a run from today, the new run is compared with the latest one.
Skip the typed confirmation with the existing ownership flag:
A longer capture, or an explicit folder:
nmap and tcpdump are required. If either is missing, the command prints one install line and stops:
On macOS, Homebrew and a pipx install often sit outside sudo's PATH:
Default passwords are not tested. That check stays opt-in, one device at a time (iotaudit credcheck, which is the only command with --yes).
auto does not ARP-spoof and does not start a hotspot. It records traffic that already passes this computer. To pull one device's traffic through this machine, use capture --mode arp below.
macOS
iotaudit scans your LAN, lists devices (IP, MAC, vendor; Tuya/Espressif flagged as likely IoT),
and lets you pick one. It then ARP-spoofs only that device so its traffic flows through your
Mac, captures for up to an hour (Ctrl+C stops early and still writes the report), and restores the
ARP tables and IP-forwarding on exit. With a device picked, --watch-lan (phone P2P) and the
IPv6/NDP filter are on by default.
Linux
Analyse an existing capture
The report lands at ./iotaudit-report.html with a matching .json next to it.
Known-issue hints
iotaudit compares banners it already has with a small catalog shipped in the
package (src/iotaudit/data/known_issues.yaml). The facts it can use are:
- an open service from the port scan, plus the nmap product and version
- an HTTP
Serverheader from a response the device sent - a UPnP or SSDP header (including an InternetGatewayDevice advertisement)
- an RTSP response that did not ask for authentication, when that was already in the capture or in the scan record
There is no lookup at runtime and no hints --update-from command. The file
changes only when you install a newer iotaudit.
Each hint is a possibility, not proof the firmware is vulnerable. The text always includes: possible, based on version string; verify firmware. The reference is the NVD page for that CVE.
Hints show in the HTML report, in the PDF (the PDF is the same HTML), in the
device JSON as known_issue_hints, and in the MCP analyze_pcap result.
Grade rule. A hint does not change the A-F grade unless both are true: the observed version string equals a version the entry lists exactly, and the entry severity is critical. That case subtracts 25 points, the same as one high finding, and the subtraction is applied after the limited-data cap so it still moves a thin capture. A version range, a product name without that exact version, an open port on its own, or any severity other than critical leaves the score unchanged. Telnet being open, an exposed UPnP IGD, or RTSP with no authentication is not by itself a CVE. The catalog cites one only when the banner or the service fact matches that entry.
Limits:
- The file is a short curated list (on the order of 10 to 20 issues), not the NVD and not a firmware database.
- A patched device can still show an old version string. An unpatched device can hide the version. Either way the hint says to verify the firmware.
- Banners are read from IPv4 packets and from the nmap record. IPv6 service banners are not parsed.
- RTSP with no authentication is recorded when a captured RTSP
200has noWWW-Authenticateheader, or when the scan extra info already says so. The shipped list does not treat that fact alone as a CVE. It is mentioned on a hint when a versioned RTSP entry also matches. - No network access is used to refresh the list.
PDF export
--pdf on analyze and capture writes a PDF next to the HTML. Print styles put the grade
and verdict at the top of each device (the first device shares page 1 with the title; every
later device starts on a new page), keep table rows in one piece, and repeat the table header.
The on-screen traffic chart is left out of the PDF.
The renderer is xhtml2pdf, a pure-Python extra (xhtml2pdf, pulled in by iotaudit[pdf]).
No Cairo or Pango install is required. If the extra is not installed, analyze and capture
still write the HTML and print one install line. iotaudit report --pdf prints that same line
and exits, because the PDF is its only output. The MCP tool analyze_pcap takes pdf=true
together with report_path and writes the PDF beside the HTML.
Compare two audits
Devices match by MAC. If either side has no MAC, they match by IP. A different MAC at the same IP is one device gone and one device new. For each device the report lists new and gone destination domains and IPs, new countries, new open ports, grade changes, verdict changes, newly seen unencrypted traffic, and new UPnP forwards when the JSON has them.
Exit code 1 when something security-relevant appeared: a device, a destination
domain or IP, a country, an open port, a worse grade, unencrypted traffic, or a UPnP
forward. Exit code 0 otherwise, including when the only changes are departures or
a better grade. A verdict change is printed either way and does not by itself exit 1.
Missing or unreadable input exits 2. With --compare, capture and analyze use
that same code after they write the report.
capture and analyze already write JSON (next to the HTML, or at --json). Keep
the previous file and pass it as --compare, or run diff on the two files later.
--json on diff prints the machine-readable result. On capture and analyze,
--json is still the path of the results file.
Cron stays quiet when nothing new showed up. This script prints the diff, and exits 1, only when something security-relevant appeared:
LAN peers are not compared. Open ports are the scanned open_ports, not the remote
port a destination was contacted on. UPnP is read from findings with category upnp,
or from a top-level upnp.mappings list joined on internal_client. An audit saved
without a UPnP check has no forwards, so the next audit's forwards show up as new.
Other findings are not listed one by one; a grade or verdict change still shows when
those findings moved the grade or the verdict.
IPv6, P2P video, and "not observed"
Live video from a camera or feeder often does not go to the cloud: it goes directly to your
phone over the local Wi-Fi (P2P), after a short STUN/TURN handshake. iotaudit tries hard to see
that stream (--watch-lan spoofs device↔phone, and the capture filter now includes IPv6 and NDP),
but two Wi-Fi clients can talk without their traffic ever passing through your computer.
because of Access-Point client isolation, a phone ignoring the spoof, or the stream riding IPv6
link-local (fe80::). When that happens the report says so honestly:
Live video set up but not observed here: the session was negotiated (STUN) but no sustained stream was captured; it most likely went straight to your phone. Not observed ≠ not happening.
For ground truth, capture at the router instead. --ipv6-ndp (experimental) also NDP-spoofs IPv6
link-local pairs.
Default-password check (opt-in)
The one active login iotaudit makes, and only when you ask for it. It tries a short list of
documented vendor default credentials (e.g. the root/xc3511 that the Mirai-era cameras
shipped with), never a password wordlist. You confirm the exact IP by typing a phrase, it only
targets a LAN address, it waits between attempts, stops at the first login that works, and reports
just "default login accepted" or "not accepted"; it never keeps or uses the session. Use
it on your own devices only.
Router port forwards (UPnP)
Many devices quietly ask the router to open a public port to themselves. iotaudit upnp lists
every UPnP port forward and the LAN device it points to. It only sends the two read-only UPnP
queries, and only to a router on a private address. It also runs after each capture (skip with
--no-upnp) and adds a finding when a forward points at the audited device. If nothing answers,
UPnP is off on your router, which is the safest setting.
Web UI and phones
View reports, upload a pcap (including PCAPdroid captures from Android) and run a one-click
scan & capture from the browser. Localhost by default; --lan is opt-in, token-protected and only
answers private-range clients. iPhones can use the web UI but cannot capture on their own; see
docs/phones.md for what works on Android and iOS and what does not.
Home Assistant
Write a small JSON summary that a Home Assistant sensor can read:
Each device gets name, ip, grade, findings, and changed (null on the first run).
Ready-to-paste sensor YAML: Home Assistant guide.
For a spreadsheet, the same summary as CSV (one row per device):
Or as a Markdown table to paste into an issue or forum post: iotaudit export audit.json --format md. Add --redact to leave out IP addresses before you post it. To share a full report in one step: iotaudit report audit.json --redact writes audit.redacted.html with local IPs, MACs and hostnames replaced by labels like lan-1. Add --pdf for a PDF (needs pip install "iotaudit[pdf]"). Or keep the JSON copy with iotaudit redact audit.json.
MCP server
diff_results compares two saved JSON audits the same way iotaudit diff does.
I also list the server on mcpservers.org and the official MCP Registry.
Consent & scope
iotaudit only audits your own network. It refuses non-private targets and requires
--i-own-this-network (or an interactive I OWN THIS NETWORK confirmation). ARP/NDP spoofing on a
network you do not own may be illegal.
Findings are indicators based on observed traffic, not proof of wrongdoing by any vendor. IP geolocation by DB-IP. MIT licensed.
來源:README.md,提交 7401939
工具
0版本歷史
17- v0.1.18最新Nov 7, 2026
- v0.1.17Nov 5, 2026
- v0.1.16Nov 4, 2026
- v0.1.15Nov 3, 2026
- v0.1.14Nov 3, 2026
- v0.1.13Nov 1, 2026
- v0.1.12Oct 30, 2026
- v0.1.11Oct 29, 2026
- v0.1.10Oct 26, 2026
- v0.1.9Oct 25, 2026
- v0.1.8Oct 24, 2026
- v0.1.7Oct 23, 2026
- v0.1.5Oct 22, 2026
- v0.1.4Oct 21, 2026
- v0.1.3Oct 20, 2026
- v0.1.2Oct 19, 2026
- v0.1.1Oct 18, 2026
評分
已滿足適用檢查項的 50%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分