SharkMCP
SharkMCP
sharkd(Wireshark의 프로그래밍 인터페이스)를 LLM을 위한 도구 세트로 노출하는 MCP 서버입니다. PCAP/PCAPNG 파일을 로드하고 자연어로 분석하세요.
요구 사항
Python 3.10+
Wireshark (
sharkd제공)
Related MCP server: Wireshark MCP
설치
git clone https://github.com/weirdmachine64/sharkmcp.git
cd sharkmcp
pip install -e .또는 설치 없이 저장소에서 직접 실행:
uvx --from git+https://github.com/weirdmachine64/sharkmcp sharkmcp구성
.mcp.json에 추가하세요:
{
"mcpServers": {
"sharkmcp": {
"command": "uvx",
"args": ["--from", "git+https://github.com/weirdmachine64/sharkmcp", "sharkmcp"],
"env": {
"SHARKMCP_TIMEOUT": "300"
}
}
}
}환경 변수 | 기본값 | 설명 |
|
| sharkd 바이너리 경로 |
|
| 요청당 타임아웃(초) |
도구
로드된 각 PCAP은 전용 sharkd 하위 프로세스를 가집니다. 비용이 많이 드는 스캔(대화, 전문가 정보, 객체 내보내기 등)의 결과는 메모리에 캐시되므로, 페이지가 나뉜 후속 호출은 재스캔 없이 처리됩니다.
세션
도구 | 설명 |
| PCAP/PCAPNG 파일 로드 |
| 로드된 모든 PCAP 목록 표시 |
| 세션 종료 및 메모리 해제 |
개요
도구 | 설명 |
| 프레임 수, 지속 시간, 파일 크기, 감지된 프로토콜 |
| 사용 가능한 모든 탭 유형, 팔로우 프로토콜, 필드 유형 |
패킷 검사
도구 | 설명 |
| 디스플레이 필터가 적용된 페이지 단위 프레임 목록 |
| 단일 프레임에 대한 전체 프로토콜 트리 |
| 패킷당 임의의 필드를 테이블로 추출 |
유틸리티
도구 | 설명 |
| 디스플레이 필터 및/또는 필드 이름 유효성 검사 |
| 접두사를 통한 필드 또는 기본 설정 이름 자동 완성 |
| 디섹터(dissector) 기본 설정 읽기 |
| 현재 세션에 대한 디섹터 기본 설정 지정 |
| 프레임 주석 달기 (세션 범위) |
트래픽 구조
도구 | 설명 |
| 프레임/바이트 수가 포함된 중첩 프로토콜 트리 |
| 간격별 프레임 및 바이트 수 |
| 다중 라인 트래픽 그래프; |
| 스트림 재조합 ( |
대화 및 토폴로지
도구 | 설명 |
| 대화 테이블 — 피어 쌍당 바이트/프레임 수 |
| 엔드포인트 테이블 — 호스트당 송신/수신량 |
두 도구 모두 지원되는 계층 유형: tcp, udp, ip, ipv6, eth, sctp, dccp, mptcp, wifi, bluetooth, zigbee, fc, fddi, usb 등.
프로토콜 통계
도구 | 설명 |
| 프레임별 이상 징후 탐지 — 오류, 경고, 참고, 채팅 |
|
|
|
|
|
|
|
|
미디어 및 VoIP
도구 | 설명 |
| 상태 및 참가자가 포함된 SIP/H.323 통화 목록 |
| RTP 스트림 인벤토리; 스트림별 지터/손실 확인을 위해 |
| UDP 멀티캐스트 스트림 통계 |
내보내기 및 객체
도구 | 설명 |
| 추출 가능한 객체 목록 ( |
| 객체, TLS 세션 키( |
탈출구
도구 | 설명 |
| sharkd 탭을 직접 실행 — 하나의 PCAP 스캔에서 최대 16개의 사양 사용 가능. 유효한 식별자를 찾으려면 |
예시
> load_pcap("/captures/traffic.pcap", alias="traffic")
> protocol_hierarchy("traffic")
> expert_info("traffic", limit=20)
> conversations("traffic", type="tcp", sort_by="bytes")
> extract_fields("traffic", ["dns.qry.name", "dns.a"], filter="dns")
> follow_stream("traffic", "http", "tcp.stream eq 0")
> export_objects("traffic", type="http")
> download_object("traffic", "eo:http_0")아키텍처
LLM
│ MCP (stdio)
▼
SharkMCP server
│ JSON-RPC 2.0 (stdin/stdout)
├─ sharkd [pcap-1]
├─ sharkd [pcap-2]
└─ sharkd [pcap-N]로드된 PCAP당 하나의 sharkd 하위 프로세스가 생성됩니다. 세션은 격리되어 있으며, 서로 다른 별칭(alias)에 대한 동시 쿼리는 서로를 차단하지 않습니다.
Available Tools
30 toolscompleteA
Autocomplete field names or preference names by prefix.
field — field prefix, e.g. "tcp." → all tcp.* dissector fields.
pref — preference prefix, e.g. "tcp." → all tcp.* preferences.
At least one of field or pref must be provided.
Dense protocols (e.g. "opcua.") can return thousands of entries;
limit caps the list (0 = no cap).
Inputs: alias (required), field, pref, limit.
conversationsB
Conversation table for a given layer. Sums frames/bytes per peer pair.
Results are sorted by sort_by (descending) and paged with skip/limit.
The PCAP is scanned only on the first call for a given filter; all
subsequent pages are served from an in-memory cache. Set limit=0 for all.
Inputs: alias (required), type, display_filter, sort_by, +2 more.
download_objectA
Download the raw content of an exportable object, TLS secrets, or RTP audio.
Tokens come from:
export_objects → "eo:http_0", "eo:imf_0", etc.
TLS session keys → "ssl-secrets" (NSS Key Log format, if embedded)
RTP audio stream → "rtp:<src_ip>_<src_port>_<dst_ip>_<dst_port>_<ssrc>"
Content is returned base64-encoded under the data key along with
the MIME type and original filename where available.
Inputs: alias (required), token (required).
endpointsA
Endpoint table for a given layer. Sums tx/rx frames and bytes per host.
Results are sorted by sort_by (descending) and paged with skip/limit.
The PCAP is scanned only on the first call for a given filter; all
subsequent pages are served from an in-memory cache. Set limit=0 for all.
Inputs: alias (required), type, display_filter, sort_by, +2 more.
expert_infoC
Per-frame expert diagnostics — errors, warnings, notes, and chats.
Returns Wireshark's built-in anomaly detection results: TCP retransmissions,
malformed packets, unusual sequences, and protocol violations.
Each entry contains frame number (f), severity (s), protocol (p),
and message (m). Results are cached after the first scan.
Inputs: alias (required), display_filter, skip, limit.
export_objectsA
List exportable objects of the given type found in the capture.
Each returned object includes a _download token that can be passed to
download_object to retrieve the raw content. Supported types:
http, dicom, smb, tftp, imf, ftp-data.
The PCAP is scanned only on the first call for a given type; subsequent pages are served from cache.
Inputs: alias (required), type, skip, limit.
extract_fieldsC
Extract arbitrary fields per packet, e.g.
["ip.src", "tcp.dstport", "http.host"]. Returns rows of dicts keyed
by field name.
Inputs: alias (required), fields (required), display_filter, skip, +1 more.
follow_streamB
Reassemble a stream. display_filter selects it (e.g. tcp.stream eq 3).
Returns client/server addrs and base64 payload chunks with direction.
Long-lived streams can produce huge payload lists; max_payloads caps
how many chunks are returned. Set to 0 for no cap.
Inputs: alias (required), protocol (required), display_filter (required), max_payloads.
get_preferenceA
Read one or all sharkd dissector preferences.
Pass a dotted preference name (e.g. "tcp.check_checksum") to read a
single value, or omit to dump all preferences. Useful for checking
whether a dissector is enabled or what port a protocol is bound to.
Inputs: alias (required), preference.
iographA
Per-interval traffic graph for up to 8 simultaneous lines.
Each entry in graphs must be one of:
"packets" | "bytes" | "bits" — basic counters
"sum:" | "avg:" — aggregate a numeric field per interval
"min:" | "max:" — min/max of a numeric field per interval
"load:" | "frames:" — bit-rate or frame count for a field
filters — optional per-graph display filters (parallel list to graphs);
enables multi-line graphs of different protocol streams.
Results are paged with skip/limit (default 300 buckets ≈ 5 min at 1 s).
Inputs: alias (required), graphs (required), interval_ms, filters, +2 more.
io_statsA
Per-interval frame and byte counts. interval_ms is the bucket size.
Results are in chronological order and paged with skip/limit.
The PCAP is scanned only on the first call for a given (interval, filter)
combination; subsequent pages are served from cache.
Default limit is 300 intervals (~5 min at 1 s buckets). Set limit=0 for all.
Inputs: alias (required), interval_ms, display_filter, skip, +1 more.
list_packetsC
Page through packets via sharkd frames.
display_filter accepts Wireshark display-filter syntax.
columns overrides the default column set with arbitrary field names
(e.g. ["frame.number", "ip.src", "tcp.dstport"]).
refs is a list of reference frame numbers for delta-time calculations.
Inputs: alias (required), display_filter, skip, limit, +2 more.
list_pcapsA
List every loaded PCAP with its current sharkd status.
Inputs: no named parameters. Returns: result
load_pcapA
Load a PCAP/PCAPNG file into a fresh sharkd session.
alias defaults to the basename. Each loaded PCAP keeps a dedicated
sharkd subprocess warm so subsequent queries are interactive.
Returns the alias and basic status (frame count, duration, file size).
Inputs: path (required), alias.
multicast_streamsB
UDP multicast stream statistics.
Returns per-stream packet counts, byte rates, and burst statistics for all UDP multicast flows in the capture.
Inputs: alias (required), display_filter.
packet_detailA
Full protocol tree for one frame.
include_bytes — add raw bytes (base64) per layer.
include_hidden — include hidden protocol tree fields.
ref_frame — reference frame number for delta-time display.
prev_frame — previous frame number for delta-time display.
Inputs: alias (required), frame_number (required), include_bytes, include_hidden, +2 more.
pcap_summaryC
Combined status + analyse: filename, size, frame count, duration,
protocols seen, first/last timestamps.
Inputs: alias (required).
protocol_hierarchyC
Protocol hierarchy stats — nested tree of frame/byte counts per protocol.
Inputs: alias (required), display_filter.
protocol_statsC
Protocol-level aggregate statistics.
protocol options: dns — query/response counts by type and return code http — HTTP request/response packet counters http_requests — requests grouped by URI http_server — load distribution across servers http_seq — HTTP request sequences http2 — HTTP/2 stream statistics rtsp — RTSP packet counters sip — SIP response code counters dhcp — DHCP message type distribution h225 — H.225 message and response status
Inputs: alias (required), protocol (required), display_filter.
response_time_delayC
Response time delay statistics for signalling protocols.
Measures per-transaction round-trip delay: h225_ras — H.225 RAS registration/admission delays megaco — MEGACO/H.248 gateway control delays mgcp — MGCP gateway control delays radius — RADIUS authentication/accounting delays
Inputs: alias (required), protocol (required), display_filter.
rtp_streamsB
RTP stream inventory and per-stream quality analysis.
Without stream_spec — returns all RTP streams with SSRC, codec,
packet count, and timing.
With stream_spec — returns jitter, packet loss, and sequence error
statistics for one stream. Format:
<src_ip>_<src_port>_<dst_ip>_<dst_port>_<ssrc>
e.g. 200.57.7.195_9762_200.57.7.196_26946_0xd2bd4e3e
Inputs: alias (required), stream_spec, display_filter.
sequence_diagramB
Flow sequence diagram data for visualising packet exchanges.
Returns time-ordered node and flow entries suitable for rendering a sequence diagram (equivalent to Wireshark's Flow Graph window). any — all flows tcp — TCP flows (handshakes, data, teardowns) icmp — ICMP echo request/reply flows icmpv6 — ICMPv6 flows
Inputs: alias (required), type, display_filter.
server_infoA
List all types available in this sharkd session: tap identifiers, follow protocols, column fields, and stats tree entries.
Useful for discovering valid values before calling conversations, endpoints, follow_stream, or protocol_hierarchy.
Inputs: alias (required).
service_response_timeC
Service response time statistics — min/max/avg latency per request type.
Measures the elapsed time between a protocol request and its response. Useful for detecting slow servers, network congestion, or anomalous response-time patterns in application protocols.
Inputs: alias (required), protocol (required), display_filter.
set_frame_commentA
Set a comment on a frame for the duration of this session (non-persistent).
Pass an empty string to clear an existing comment. Useful for annotating frames during forensic analysis.
Inputs: alias (required), frame_number (required), comment.
set_preferenceA
Set a sharkd dissector preference for this session.
Changes are session-scoped and do not persist after the server restarts.
Common uses: forcing a non-standard port to decode as a specific protocol
(e.g. name="http.tcp.port", value="8080"), or enabling/disabling a
dissector option.
Inputs: alias (required), name (required), value (required).
tapA
Run one or more sharkd statistics taps in a single PCAP scan.
specs — tap identifiers from server_info, e.g.
["expert", "conv:TCP", "stat:dns", "srt:smb"]
filter — global display filter applied to all taps (sharkd supports
only one filter per tap call; per-tap filters are iograph-only)
skip / limit — pagination applied to each tap's flat list result
Up to 16 specs per call (sharkd limit). Results are cached after the
first scan; paginated follow-up calls are served from memory.
Use server_info to discover all valid tap identifiers.
Inputs: alias (required), specs (required), filter, skip, +1 more.
unload_pcapB
Terminate the sharkd session for alias and free its memory.
Inputs: alias (required).
validateA
Validate a display-filter expression and/or a field name.
filter — Wireshark display filter (e.g. tcp.port == 443).
field — fully qualified field name (e.g. http.request.method).
Returns {"status":"OK"} on success or an error object.
At least one of filter or field must be provided.
Inputs: alias (required), filter, field.
voip_callsC
VoIP call list with state, duration, and participant addresses.
Detects SIP, H.323, MGCP, and SKINNY calls in the capture and reconstructs the call flow from signalling messages.
Inputs: alias (required), display_filter.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 30 tools
Each tool targets a distinct operation on packet capture data, such as loading, listing, extracting fields, following streams, or computing statistics. Overlaps like io_stats and iograph are clarified by descriptions (one simple counts, the other multi-line graphs). No two tools appear to do the same thing.
Tool names mix verb_noun patterns (list_pcaps, load_pcap, set_preference) with noun-only names (protocol_hierarchy, voip_calls, rtp_streams). While the naming is readable and uses snake_case consistently, the lack of a uniform verb prefix creates inconsistency.
30 tools cover a broad range of PCAP analysis features (loading, filtering, statistics, protocol details, export). This is slightly above the typical 15-20 but justified given the extensive functionality of Wireshark/sharkd. No obvious bloat.
The tool surface is comprehensive for analyzing packet captures: it covers file management, packet listing, field extraction, stream reassembly, export objects, multiple statistics taps, protocol-specific analysis, preferences, and validation. Missing features like live capture or saving are outside the server's stated purpose.
Maintenance
Related MCP Connectors
Use your Mac, Windows or Linux computer from ChatGPT, Claude or Codex: files, commands, documents.
File tools for your AI chat: PDF tables to Excel, merge PDFs, images to PDF, fill PDF forms.
File uploads for AI agents. Upload, list, and manage files. No signup required.
Open, inspect, filter, edit and convert xlsx and csv files from your AI chat. Processing is local.
Related MCP Servers
- AlicenseCqualityDmaintenanceEnables LLMs to perform automated network forensics, threat hunting, and traffic analysis using Wireshark/TShark.344MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to analyze network traffic using Wireshark/tshark, providing packet statistics, protocol analysis, and anomaly detection through natural language interaction.75MIT
- FlicenseCqualityDmaintenanceEnables LLMs to capture, analyze, and summarize network traffic using Wireshark CLI tools, supporting live capture, pcap analysis, and LLM-oriented summaries.76-
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to interact with tshark and related command-line tools for live packet capture, PCAP analysis, filtering, stream following, and capture file merging.1MIT