netimps

The network utilities every tool ends up rewriting — interface discovery, "which is my IP", reachability checks, CIDR maths, host:port parsing, DNS, ping, scanning and multicast — as one typed, flat-import library.

Built on the standard library, with no hard runtime dependencies. resolve() tries dnspython, the OS resolver and nslookup, in that order, using whichever is available (pip install netimps[dns] for dnspython itself, the fullest of the three: structured records, every record type, explicit nameserver/search-list control). Interface enumeration, routing and multicast are ctypes bindings to the platform's own APIs, so there is nothing to compile and no wheel to miss for your platform.

Features

• Interface discovery, no dependencies — get_interfaces() gives adapter names, MACs, MTU and real prefix lengths on Linux, macOS/BSD and Windows, via getifaddrs(3) / GetAdaptersAddresses. No ifaddr required.
• One parsing entry point — parse(value, type) with non-raising try_parse and boolean is_valid siblings. Type-form overloads preserve the parsed result, including v4/v6 union aliases and caller-supplied defaults.
• MACAddress — colon/hyphen/dot/bare plus int/bytes, hashable and ordered, with .oui, .is_multicast, .is_local and case-selectable rendering.
• The socket helpers everyone rewrites — get_source_ip, get_free_port, tcp_check, wait_for_port.
• Local membership lookups — interface_for() gives the first adapter for an address, interface, network or MAC; interfaces_for() yields every match, and is_local_address() distinguishes assignment from mere address scope.
• Routing and MTU — get_route (first hop, unprivileged), hop_count (raw sockets or traceroute fallback), discover_mtu / get_pmtu, Interface.mtu.
• CIDR set maths — collapse and subtract, the latter missing from ipaddress entirely.
• normalize_host — host:port splitting that gets IPv6 brackets right.
• Scanning — concurrent scan_ports / scan_hosts.
• Socket setup — bind() with the options named, UdpEndpoint for UDP servers that need to know which interface a datagram arrived on, and retry() for the backoff loop everyone writes without jitter.
• Multicast — multicast_socket handling the join dance whose failure modes are otherwise silent.
• DNS and ping — resolve() chaining dnspython/OS resolver/nslookup backends and returning native types, accepting an IP/interface object directly and auto-selecting a reverse lookup for an address query; ping() returning round-trip time and TTL, not just a boolean.

Installation

pip install netimps          # the library -- no hard dependencies
pip install netimps[dns]     # plus dnspython, for resolve()'s fullest backend
pip install netimps[cli]     # plus the `netimps` command

Requires Python 3.9+. Both extras are additive and independent; importing the library never requires either.

Command line

netimps interfaces                       # names, MACs, MTU, addresses
netimps ping 8.8.8.8 -m tcp -p 443       # icmp | tcp | udp
netimps resolve example.com aaaa
netimps resolve 8.8.8.8                  # no rdtype -> auto ptr -> dns.google
netimps check example.com https          # port number or scheme name
netimps mtu 8.8.8.8                      # measured, not guessed
netimps scan 192.0.2.0/29 -p common
netimps addr 00:00:5e:00:53:01           # address, network or MAC
netimps split '[::1]:8080'               # -> ::1  8080

Every command takes --json. Exit codes are meaningful: 0 success, 1 "the answer was no", 2 a caller error.

Quick start

import netimps
from netimps import IPAddress, IPNetwork, MACAddress, parse

# Interfaces: names, MACs, MTU and real prefixes on every OS
for iface in netimps.get_interfaces():
    print(iface.name, iface.mac, iface.mtu, [str(ip) for ip in iface.ips])
    # 'Wi-Fi'  00:00:5e:00:53:01  1500  ['192.0.2.10/24', 'fe80::.../64']

# Types annotate; parse builds
def route(dst: IPAddress, via: IPNetwork) -> None: ...

parse("10.0.0.5")                        # IPv4Address('10.0.0.5')
parse("10.0.0.5/24", IPNetwork)          # IPv4Network('10.0.0.0/24')
netimps.try_parse("nope", IPAddress)     # None
netimps.is_valid("::1", IPAddress)       # True

# Which of my addresses actually reaches that host?
netimps.get_source_ip("8.8.8.8")         # IPv4Address('192.0.2.10')
netimps.get_route("8.8.8.8").gateway     # IPv4Address('192.0.2.1')

# Exact assignment, subnet membership, or MAC ownership
netimps.interface_for("192.0.2.10")      # first matching Interface, or None
list(netimps.interfaces_for(parse("192.0.2.0/24", IPNetwork)))
netimps.is_local_address("127.0.0.1")    # True without interface discovery

# Honest reachability, and waiting for a service
netimps.tcp_check("example.com", 443)              # True
netimps.wait_for_port("localhost", 5432, timeout=60)

# CIDR set maths
netimps.subtract(["10.0.0.0/24"], ["10.0.0.64/26"])
# [IPv4Network('10.0.0.0/26'), IPv4Network('10.0.0.128/25')]

# host:port, including the IPv6 case people get wrong
netimps.normalize_host("[::1]:8080")     # ('::1', 8080)
netimps.normalize_host("::1")            # ('::1', None)  -- not port 1

# MAC addresses
mac = MACAddress("AA-BB-CC-DD-EE-FF")
mac.as_str("-", upper=True)              # 'AA-BB-CC-DD-EE-FF'
mac.is_local, mac.oui.hex()              # (True, 'aabbcc') -- AA has the U/L bit

# DNS returns native types
netimps.resolve("example.com")[0].is_global   # an IPv4Address, not a str
netimps.resolve("example.com", "txt")         # ['v=spf1 -all']  -- unquoted

# ping carries the details, and can use TCP or UDP where ICMP is blocked
result = netimps.ping("8.8.8.8")
result.ok, result.rtt_ms, result.ttl     # (True, 9.0, 119)
netimps.ping("8.8.8.8", method="tcp", port=53)   # times the handshake

# Path MTU, measured rather than guessed
netimps.discover_mtu("8.8.8.8")                       # 1500
netimps.discover_mtu("10.0.0.5", method="udp", port=9999)
netimps.get_tcp_mss("example.com", 443)               # 1460

# Scanning and multicast
netimps.scan_ports("192.168.1.1", ["ssh", "https"])   # [22, 443]
sock = netimps.multicast_socket("224.0.0.251", 5353)  # mDNS listener

# Server-side: bind with the options named, and know where packets came from
server = netimps.bind("", 6767, broadcast=True)
endpoint = netimps.UdpEndpoint(server)

# Retry with backoff and jitter
netimps.retry(lambda: netimps.tcp_check("example.com", 443), attempts=3)

API overview

Full per-export reference, with contracts and gotchas, lives in src/netimps/AGENTS.md.

Design notes

A few behaviours are deliberate and worth knowing:

• Interface.is_loopback is computed from addresses, not names — lo, lo0 and Loopback Pseudo-Interface 1 share no spelling.
• Concrete types are strict about family. parse("::1", IPAddress) works; parse("::1", IPv4Address) raises rather than quietly returning v6.
• Networks parse non-strict by default, so 10.0.0.5/24 normalises instead of raising. Pass strict=True for stdlib behaviour.
• Addresses can belong to more than one interface. interface_for() keeps the historical first-match result; use interfaces_for() when duplicates such as unscoped IPv6 link-local assignments matter.
• resolve raises on a malformed query rather than returning [] — a typo'd record type should not look like "no such record".
• ping(ttl=...) behaves the same on every OS. Windows ping exits 0 for "TTL expired in transit", so the reply address is verified instead of trusting the exit code.
• hop_count works unprivileged, falling back to the system traceroute when a raw socket is unavailable.
• discover_mtu measures; get_pmtu only reports what the kernel cached. The latter is usually None, and always None on Windows. On one real host the local link was 9000 and the true path MTU 1500 — only probing found it.

Development

python -m venv .venv/dev
.venv/dev/Scripts/pip install -e ".[dev]"   # POSIX: .venv/dev/bin/pip
.venv/dev/Scripts/pytest -q
.venv/dev/Scripts/black src/ tests/

Tested on Python 3.9 (the floor) and 3.14.

Releasing

This project follows Semantic Versioning and keeps a CHANGELOG.md. Pushing a tag matching v* triggers the release workflow: test gate → build → publish → docs deploy.

License

MIT — see LICENSE.