v1.7.1

On PyPI now, and in the official MCP registry.

pipx install proxy-scraper-cli

The plain name proxy-scraper is blocked on PyPI by an older, similar package, so the package is proxy-scraper-cli – the commands stay proxy-scraper and proxy-scraper-mcp, imports stay proxyscraper.

The MCP server is listed as io.github.maximilianfeix/proxy-scraper on registry.modelcontextprotocol.io, and every release updates it by itself. Claude Code:

claude mcp add proxy-scraper -- uvx --python ">=3.10" --from "proxy-scraper-cli[mcp]" proxy-scraper-mcp

proxy-scraper --mcp starts the MCP server as well (#42).

v1.7.0

Your AI agent can use the proxies now.

New

• MCP server – proxy-scraper-mcp for Claude Code, Claude Desktop, Cursor, VS Code, Codex and other MCP clients (#129):

  • get_proxies – working proxies instantly, from the hourly list, with the website's filters
  • check_proxies – checks from your own network, reports progress while it runs
  • fetch_url – loads a page through a verified proxy as readable text, e.g. "load bbc.co.uk/news as seen from the UK". Switches proxies when one fails, HTTPS only through proxies with verified TLS, and a proxy is only blamed when another one gets the page – a site with an expired certificate doesn't cost you proxies. Local and private addresses are refused, also after redirects and for names that resolve to them.

  Claude Code, one line:

  claude mcp add proxy-scraper -- uvx --python ">=3.10" --from "proxy-scraper[mcp] @ git+https://github.com/maximilianfeix/proxy-scraper" proxy-scraper-mcp

  Other clients: README → For AI agents

• More sources – 28 new lists, and the GitHub search runs daily now and keeps what it found instead of starting over every 3 days (#127)

Fixed

• Proxy server: stopping it ends open tunnels (on Python 3.12+ Ctrl+C could hang on one), IPv6 targets in plain http:// URLs
• Website: calmer trend chart, footer links no longer overlap the wordmark, less empty space (#131)

Next

PyPI (#42) – which also gets the MCP server into the official MCP registry – and the Hacktoberfest issues. See the v1.8 milestone.

Full list: CHANGELOG.md · v1.6.0...v1.7.0

v1.6.0

The list got faster and easier to find, the proxy server can run on its own for days, and everything is English now.

New

• Blocklist check – every exit IP is looked up on SpamCop, with --no-blocklisted to skip listed ones. In today's list 28 % are listed. If your DNS resolver gets refused it says so instead of guessing (#66)
• Live list every hour – instead of every 6 hours, a full run only takes about 3 minutes. "Stable" still means 24 h (#111)
• Pages per protocol and country – socks5/, country/de/ and so on, each with a proxies.txt, plus a sitemap (#121)
• Password for the proxy server – PROXY_SCRAPER_SERVE_PASSWORD or --serve-password, for HTTP, SOCKS5 and the status page. Plus a compose.yaml that runs it in Docker (#103)
• The server refills itself – --serve-refill 6 checks fresh proxies in the background every 6 hours and adds the hits, so a long-running server doesn't run dry (#109)
• -o - prints the hits to stdout for pipes, the interface moves to stderr (#105)
• Discord bot in bot/ – posts the live list, answers /proxies, /proxy, /stats (#74, #75)
• New website and logo, README redesign, WCAG fixes on the site, and the whole repo is in English now (#71, #72, #73, #79, #80)

Fixed

• A target that's down no longer gets working proxies disabled in the proxy server, IPv6 targets work over CONNECT and SOCKS5 (#82)
• Proxies still being confirmed when --want is reached no longer count as dead, outdated sources get another look every 3 days (#83)
• A handful of smaller robustness bugs (#84)

Next

PyPI (#42), protocol detection on the same connection (#44), and a few Hacktoberfest issues if you want to help. See the v1.7 milestone.

pipx install git+https://github.com/maximilianfeix/proxy-scraper.git
# or
docker run --rm ghcr.io/maximilianfeix/proxy-scraper --want 20

Full list: CHANGELOG.md · v1.5.0...v1.6.0

v1.5.0

Big one. The proxy server grew up, the live list got a website, and there's a Python API now.

New

• Live list website – https://maximilianfeix.github.io/proxy-scraper/ with search, filters, copy/download, country and latency charts and a trend over the last runs (#47)
• Stable proxies – the list remembers how many runs in a row each proxy worked. "Stable 24 h+" filter and an "Up for" column on the website (#61)
• Tampering check – a static page has to arrive byte for byte. Out of 270 working proxies, 54 injected something, mostly a <script> tag. Those are gone now (#46)
• Provider and datacenter – ASN and provider name for every exit IP (offline, DB-IP), plus --no-datacenter. About 45 % of working proxies exit from datacenters (#49)
• Proxy server – --rotate weighted|random|round-robin|fastest, --sticky SEC, choosing per request through the username (country-de-type-socks5-session-abc), SOCKS5 on the same port, dead proxies get re-checked every 5 minutes, --serve-host for Docker (#48, #43)
• Status and metrics – /__proxy-scraper/status as JSON and /__proxy-scraper/metrics for Prometheus/Grafana (#67)
• --recheck live – starts from the public list and checks it again from your own network. 517 of 1,169 worked, in 31 s instead of a full scan (#60)
• Python API – from proxyscraper import find_proxies, same checks as the CLI (#50)
• Shell completion – eval "$(proxy-scraper --completion zsh)", also bash and fish (#62)

Next

PyPI (#42), blocklist info for exit IPs (#66, measured: 29 % are on SpamCop) and protocol detection on the same connection (#44), see the v1.6 milestone.

pipx install git+https://github.com/maximilianfeix/proxy-scraper.git
# or
docker run --rm ghcr.io/maximilianfeix/proxy-scraper --want 20

Full list: CHANGELOG.md · v1.4.0...v1.5.0

v1.4.0 – exports, logins, Docker and a lot less waiting

Worked through all open issues for this one.

New

• Exports – --export proxychains,clash,curl (or all) writes ready-to-use configs next to the results. HTTP proxies that can't tunnel are left out of proxychains and Clash, since both only use CONNECT (#4)
• Proxies with credentials – socks5://user:pass@… entries keep their login: Basic auth for HTTP, RFC 1929 for SOCKS5, user ID for SOCKS4. They're never published to the public live list, and the terminal only shows user:••• (#7)
• Fallback check targets – five targets instead of one, none behind Cloudflare. If the current one goes down mid-run, the tool switches and re-checks what was affected, so the source stats don't learn from an outage (#8)
• Docker image – ghcr.io/maximilianfeix/proxy-scraper for amd64 and arm64 (#5)

Faster

• ETag cache – unchanged lists aren't downloaded again. A second run right after the first: 0 MB in 9 s instead of 161 MB in 22 s (#9)
• Countries offline – DB-IP Lite instead of waiting for ip-api.com at the end of a run (#2)

Measured and dropped

• Protocol detection with an extra connection made runs slower (77 s vs 60 s), see #3
• IPv6: only ~300 entries in all 726 sources, mostly V2Ray configs, see #1

pipx install git+https://github.com/maximilianfeix/proxy-scraper.git
# or
docker run --rm ghcr.io/maximilianfeix/proxy-scraper --want 20

Full list: CHANGELOG.md · v1.3.0...v1.4.0

v1.3.0 – installable, polished, English

Install

pipx install git+https://github.com/maximilianfeix/proxy-scraper.git
proxy-scraper

Added

• Installable command-line tool – pipx install … gives you proxy-scraper, plus python -m proxyscraper and --version. Learned state moves to your user data folder, results go to ./results (#26)
• Release workflow – every version tag is tested, built, smoke-tested and published with a wheel
• Next steps after a run: test the fastest proxy with curl, start the proxy server, recheck later (#27)
• English README (German in README.de.md), contributing guide, security policy, code of conduct, issue forms (#28)

Changed

• Calmer, consistent terminal UI: one colour palette, framed setup section, clearer panel titles and dashboard cards (#27)
• Live list badges in English

Full changelog: CHANGELOG.md · v1.2.0...v1.3.0

v1.2.0

Neu

• Einrichtungsassistent – ohne Argumente gestartet fragt das Tool per Pfeiltasten, was du suchst: Schnellauswahl (Surfen & Web, Maximal anonym, Sofort als Proxy-Server …) oder Schritt für Schritt. Zeigt am Ende die passende Kommandozeile und merkt sich die letzte Auswahl (#12)
• Rotierender Proxy-Server – --serve stellt die Treffer auf 127.0.0.1:8899 bereit; jede Verbindung über einen anderen Proxy, unbemerkter Wechsel bei stummen oder falschen Antworten, für HTTPS nur Proxys mit verifiziertem TLS (#15)
• Zielseiten – --target google.com behält nur Proxys, die die Seite wirklich erreichen (#14)
• Live-Proxyliste – alle 6 Stunden per GitHub Actions auf dem Branch proxy-list, mit Badges in der README (#18)

Besser

• Honeypots werden aussortiert – jeder Treffer muss eine zweite, unabhängige Seite mit derselben Exit-IP abrufen. In einem Testlauf waren vorher 699 von 754 „Treffern“ Fakes (#20)
• Schneller mit Filtern – --max-latency wird zum Timeout, überflüssige HTTPS-Tests entfallen (bei langsamen Proxys 3× schneller) (#13)
• aufgeräumte Architektur: RunOptions, Ablauf in Phasen, UI als Paket (#11)
• neue README mit animierter Demo, Lint (ruff), CodeQL, Dependabot

Zahlen

233 Tests (vorher 54), CI auf Linux, macOS und Windows mit Python 3.9 / 3.11 / 3.13.

v1.0.0

Erste veröffentlichte Version.

Was drin ist

• über 700 Quellen: kuratierte Liste, Meta-Quellen und automatische Suche auf GitHub
• eigene Handshakes für HTTP, SOCKS4 und SOCKS5, 2000+ Prüfungen parallel
• HTTPS-Test, Anonymitätsstufe und Land für jeden Treffer
• lernt pro Quelle die Trefferquote und prüft bekannte Proxys zuerst
• Filter (--country, --https-only, --anonymity, --max-latency) und --want N
• Ergebnisse als txt, json und csv unter results/
• Live-Dashboard im Terminal

Bekannt

• läuft noch nicht unter Windows (#6)
• in Netzen mit strenger Firewall kommt kaum etwas durch – das Tool warnt dann