HProxy · Free South Africa pool
Free South Africa proxy list 2026
Every South African row of our list went through two test rounds. Each relay stayed in the country.
Every slice of the pool
The same live list, one thing at a time.
A page per country, protocol, anonymity grade, operator, port, network and block, each with its own count, download and API call.
South Africa proxies right now, by the numbers
The pool currently holds 367 South Africa proxies spread across 39 different networks. They respond in 2161ms at the median, slower than the 1530ms median across the whole free pool. The busiest exits are Cape Town, Brits, Johannesburg. These figures come straight from the live pool and move every few minutes as proxies are re-verified.
By anonymity
Biggest hosting networks
About
What our tests found on the South African list
These are public proxies gathered from across the web and re-verified continuously by HProxy's engine. They are free, and they are fine for quick, low-stakes tasks. Like all public proxies they are shared and they go down without warning. Most of them are already blocked by the big sites. Always test one before you rely on it.
Verify before you use
Run any list through the Proxy Checker, in the free app or online, and see which proxies are alive, fast and anonymous right now.
The long read
About free South African proxies.
Thirty-three South African entries were live in our pool on 28 September 2026, at 09:52 UTC.
All but one carried no datacenter flag: they sat on South African networks such as MTN and SEACOM.
Every relay in our tests left from inside South Africa: 47 of 47 over two passes.
Amazon rows in Cape Town, 48 of the 154 on the 48-hour list, relayed nothing.
At 11:13 UTC they were also slower than the pool's ordinary entries and passed fewer checks.
For work that needs the same address every day, paid South Africa proxies are the better fit.
01
Local networks, weaker checks
At 09:52, the South African slice barely touched datacenters: one entry of 33.
By 11:13 UTC, 20 entries were live, and the table sets them against the pool's ordinary addresses.
| Measured 28 September 2026 | South Africa | Rest of the pool |
|---|---|---|
| Live entries measured | 20 | 3,881 |
| Median success rate across its own checks | 25.2% | 33.9% |
| Passed ten checks at 90 percent | 0.0% | 9.4% |
| Median latency from our server in the United States | 7,162 ms | 6,094 ms |
| Flagged as datacenter by our list | 5.0% | 21.4% |
Read at 11:13 UTC; both columns leave out addresses holding 20 or more live entries.
At 09:52, HERO Telecoms carried six live entries, Info-Gro five, MTN four and SEACOM three.
SOCKS4 carried 10 of those 33, and not one offered SOCKS5.
Nine were graded transparent, which means a site could see the address behind them.
02
Every relay stayed in South Africa
ip-api.com told our server which address each relay arrived from, and where that address sits.
| Measured 28 September 2026 | Entries |
|---|---|
| Tested | 32 |
| Did not accept a connection | 17 |
| Relayed the plain web request | 11 |
| Of those, exiting in South Africa | 11 |
| Of those, leaving from the listed address | 9 |
| Opened Google over an encrypted connection | 6 |
| Certificate did not verify | 0 |
Our United States server ran this test at 09:53 UTC, with ip-api.com placing every exit.
The 11 relays took from 0.9 to 12.8 seconds, about 5.6 at the median.
The other two left from a second South African address, in Randburg and Durban.
SOCKS4 rows relayed 5 of 10, and HTTP and HTTPS rows 3 of 11 each.
On the 48-hour list, the rows off Amazon gave 17 relays, all inside South Africa.
Seven rows reached Cloudflare over HTTPS, and Cloudflare placed every one in South Africa.
In our second pass at 10:34, just 5 of the 32 live entries relayed, all inside South Africa.
The list rows off Amazon then gave 14 relays, again every one in the country.
03
Amazon rows in Cape Town
At 09:54 UTC, six Amazon addresses in Cape Town held 48 of the 154 list rows.
ARIN registers the three blocks they sit in to Amazon Data Services South Africa.
None of the 48 relayed: 42 answered HTTP 400 and 6 answered 501.
Twenty of the 48 also showed certificates that failed verification on encrypted requests.
At 10:34, four of those addresses carried 36 rows, and again none relayed.
Of 1,608 South African entries that worked and then stopped, 1,385 sat on Amazon (17:17 UTC, 28 September).
When we read another free list at 09:59, 17 of its first 20 rows sat on these blocks.
04
Lifespans a little shorter than the pool's
On ordinary addresses, 2,060 of 2,293 retired entries never worked: 89.8 percent, against 88.6 elsewhere.
Amazon entries, which relayed nothing for us, stay out of the lifespans below.
| Measured 28 September 2026 | South Africa | Rest of the pool |
|---|---|---|
| Worked, then stopped | 1,608 | 296,295 |
| Left out: on addresses with 20 or more | 1,375 | 260,277 |
| Left out: the evening of 14 June | 18 | 9,174 |
| Left out: other entries on Amazon's networks | 10 | 1,056 |
| Counted | 205 | 25,788 |
| Median lifespan of those counted | 8.5 days | 9.8 days |
| Of those counted, stopped within a day | 24.4% | 24.7% |
| Of those counted, lasted more than thirty days | 29.3% | 31.0% |
Read at 17:17 UTC. On the June evening, a fault in our checker marked about 128,000 entries dead at once.
The typical entry stopped about a day sooner here, and slightly fewer lasted past a month.
05
No confirmed blocks in OONI's counts
OONI's tests from South Africa found no confirmed block in the month to 28 September.
Of the 172,351 web tests in that month, 96.2 percent loaded as they normally do.
YouTube, Facebook and Instagram each loaded normally in about 99 percent of their tests.
06
Putting a South African row to work
Leave out the rows on Amazon's Cape Town blocks, 13.244 to 13.246, 16.28 and 15.240.
Off those blocks, rows showing 50 percent uptime or more relayed 5 of 10, then 3 of 10.
Rows below 50 percent relayed 12 of 96, then 11 of 96, not even half as often.
Among HTTP, HTTPS and SOCKS4, the SOCKS4 rows relayed most often in all four of our tests.
Retry a failed row before you drop it.
Of the live rows that relayed, 11 of 16 did so only once, then 5 of 8.
In the 09:52 read, 8080 served ten live entries, and 4153 and 5678 four each.
Test a row in our proxy checker, which shows its speed, anonymity, country and network.
Questions
Frequently asked.
Yes, in our tests: all 47 relays over two passes left from inside South Africa. Most of them came from the very address the list showed.
Slower than the pool: at 11:13 UTC, half the live entries took over 7,162 ms to answer. On the pool's other ordinary addresses, half answered within 6,094 ms, from our server in the United States.
No. Nine of the 33 live entries were transparent, handing each site the address behind them. On encrypted requests, the Amazon rows also showed certificates our server could not verify: 20, then 6.
Half the 205 entries we could time were gone within 8.5 days, against 9.8 days across the pool. Neither count includes addresses holding 20 or more such entries, Amazon's networks or one faulty June evening.
Most rows on the 48-hour list had stopped: 74 of 106 off Amazon took no connection, then 80. The 48 Amazon rows, then 36, took one and relayed nothing.
Johannesburg held 9 of the 33 live entries and Cape Town 3. The town beside a row comes from a location database, so treat it as a guess.
Upgrade to the paid network
Need pools that stay clean?
Free IPs are shared and rotate out fast, so success rates swing. The paid network gives you dedicated-clean pools that hold up under load.
Docs and AI
The free list, in your assistant and in your code.
We built the MCP server and the CLI ourselves: the free proxy list is one tool call away in Claude, ChatGPT, Cursor or any assistant that speaks MCP, and the list's API is documented end to end.
Every product has its own reference in the docs. Something the list does not give you? Tell us.
Claude, ChatGPT, Cursor, Gemini, Copilot, Windsurf, Perplexity, Ollama and any assistant that speaks MCP.
MCP server
Built by us and hosted at mcp.hproxy.com: the free proxy list, the checker and the IP lookup as tools in Claude, ChatGPT, Cursor, VS Code or any assistant that speaks MCP. Nothing to install, no key.
CLI
The list from a terminal: hproxy list --country DE --limit 20, from the command line of the free HProxy app.
API docs
The free list's API, documented end to end with real responses: the filters, the formats, the limits.
llms.txt
The whole documentation as markdown for any model that reads: /llms.txt for the index, /llms-full.txt for everything.
Let your assistant pull the list
The pool is a free, keyless API and a documented markdown page. Hand your assistant one of these and it can fetch live proxies on its own, filtered the way you asked.
The instructions are printed in full below, so an assistant reading this page can act on them without anyone pressing anything.
All the machine lanes- MCP server
- Adds proxy_list, proxy_check and ip_lookup as tools your assistant can call by itself. Nothing to install.
claude mcp add --transport http hproxy https://mcp.hproxy.com/mcpClaude, ChatGPT, Cursor, VS Code, Windsurf or any assistant that speaks MCP: add https://mcp.hproxy.com/mcp as a remote server. This page's tool is proxy_list. - Command line
- The same tool without an assistant, from the free HProxy app's command line. Add --json for JSON.
hproxy list --country DE --protocol socks5 --limit 20Comes with the free HProxy app: hproxy.com/proxy-checker - REST, no key
- Plain HTTP with CORS on, which is all an agent with a fetch tool needs.
curl "https://hproxy.com/api/proxy-list?format=json&country=de&limit=20"
Instructions for an assistant, in full
Everything a model needs to call the free proxy list without an account: the endpoint, the parameters, the limits and the honest caveat. Readable straight off the page.
You can use HProxy's free proxy list, which needs no API key and no account.
GET https://hproxy.com/api/proxy-list?format=json returns the live pool as JSON. Useful query parameters: country (ISO alpha-2, e.g. de), protocol (http, https, socks4, socks5), anonymity (elite, anonymous, transparent) and limit. Add format=txt for plain ip:port lines instead.
Each proxy comes back with ip, port, protocols, anonymity, country_code, city, asn_org, latency_ms and uptime_24h. The pool is re-checked around the clock, so ask for it fresh rather than caching it.
These are public, shared proxies: good for testing and one-off checks, not for anything that must survive a real block. Full docs: https://hproxy.com/llms/free-proxy.txt
There is also an MCP server, so an assistant can call this as a tool instead of writing HTTP: add https://mcp.hproxy.com/mcp as a remote MCP server (nothing to install, no key) and use the proxy_list tool. In Claude Code the line is `claude mcp add --transport http hproxy https://mcp.hproxy.com/mcp`. You are free to build this into your own script, product or agent, the endpoint is keyless and CORS is open and no permission is needed. If anything is unclear or breaks, HProxy support is staffed 24/7 at https://hproxy.com/contact.Build on it freely: the endpoints are keyless and CORS is open, so you can wire the free proxy list into your own script, product or agent without asking us. If you get stuck, support answers around the clock, every day, at hproxy.com/contact. Generated markdown doc: hproxy.com/llms/free-proxy.txt.
About this list & IP removal
Every proxy here is aggregated from publicly available sources. We do not scan or collect these IP addresses ourselves: we verify entries that are already public (live status, latency, uptime) and store nothing else about the devices behind them. The list is provided as-is, with no warranty, for lawful use only, on the basis of our legitimate interest in a free, transparent proxy directory (Art. 6 (1) (f) GDPR). Never route passwords or sensitive data through a public proxy. If an IP address is yours and you'd like it removed, email legal@hproxy.com and we'll take it down (Art. 21 GDPR). Full details on the report abuse page.