PacketMaster
MCP server for network-troubleshooting PCAP analysis via tshark, enabling users to analyze PCAP files, detect anomalies, and troubleshoot network issues.
README
PacketMaster
MCP server for network-troubleshooting PCAP analysis via tshark. Pairs with Cursor skills pcap-troubleshooting (domain workflows) and packetmaster-mcp (MCP tool usage).
Prerequisites
- Python 3.10+
- uv (recommended) or pip
- Wireshark with CLI tools:
tshark,capinfos
brew install wireshark # macOS
PacketMaster auto-detects `/Applications/Wireshark.app/Contents/MacOS/tshark` if not on PATH.
Install
cd "/Users/networkze/Documents/Claude Projects/PacketMaster"
uv sync --extra dev
Cursor MCP Configuration
Install the launcher (once), then add to ~/.cursor/mcp.json:
mkdir -p ~/.packetmaster-mcp
cp scripts/start-mcp.sh ~/.packetmaster-mcp/start-mcp.sh
chmod +x ~/.packetmaster-mcp/start-mcp.sh
{
"mcpServers": {
"packetmaster": {
"command": "/Users/networkze/.packetmaster-mcp/start-mcp.sh",
"env": {
"FASTMCP_SHOW_SERVER_BANNER": "false"
}
}
}
}
Uses a launcher script (like flipper-zero) because Cursor can break uv run paths that contain spaces.
PCAP paths must be absolute. By default any readable .pcap on disk is allowed. Set PM_ALLOWED_DIRS only if you want an opt-in sandbox.
Environment Variables
| Variable | Default | Description |
|---|---|---|
PM_ALLOWED_DIRS |
(unset) | Optional colon-separated allowlist; when set, paths must stay inside |
PM_REQUIRE_ALLOWED_DIRS |
false | If true, PM_ALLOWED_DIRS must be set or startup validation fails |
PM_MAX_FILE_SIZE |
10GB | Max PCAP file size |
PM_MAX_PACKETS |
10000 | Max packets per extraction request |
PM_MAX_STDOUT_BYTES |
10MB | Max subprocess stdout |
PM_TIMEOUT |
300 | Subprocess timeout (seconds) |
PM_HASH_MAX_BYTES |
500MB | Skip SHA-256 above this size unless requested |
PM_REDACT_SENSITIVE |
true | Redact credentials in stream/field output |
TSHARK_PATH |
auto | Override tshark binary path |
Tools
| Tool | Description |
|---|---|
pm_check_install |
Verify tshark/capinfos installation |
pm_file_info |
capinfos + optional SHA-256 |
pm_protocol_hierarchy |
Protocol distribution |
pm_conversations |
Top conversations by bytes |
pm_endpoints |
Top IP endpoints |
pm_expert_info |
Wireshark expert information |
pm_tcp_anomalies |
TCP retrans/dup ACK/zero window/OOO/RST |
pm_latency_summary |
Per-flow RTT statistics |
pm_io_stats |
Traffic over time |
pm_extract_fields |
Paginated field extraction |
pm_follow_stream |
Follow TCP/UDP/HTTP/TLS stream (redacted) |
pm_filter_packets |
Packet summaries by display filter |
pm_detect_capture_type |
Start here — wlan vs ethernet + tool routing |
pm_wired_quick_scan |
EPC/SPAN — DHCP, DNS, ICMP, VLAN, L2 infra, TCP |
pm_dhcp_analysis |
DHCP phases, stuck patterns, optional MAC filter |
pm_dns_analysis |
DNS queries/responses, NXDOMAIN, top names |
pm_asymmetric_hints |
One-sided capture / path asymmetry heuristics |
pm_troubleshoot_quick_scan |
L3 TCP-focused composite (alternate to wired scan) |
pm_troubleshoot_flow |
Deep dive between two IPs |
pm_wifi_quick_scan |
802.11 monitor — retries, mgmt, roaming, BSSIDs |
pm_wifi_client_analysis |
Deep dive on one STA by MAC |
pm_wifi_roaming_analysis |
802.11k/v roam timeline for one STA |
Workflow with NetOps EPC
- Capture via
ios_xe_epc_capture_run(user-netops MCP) withpull_pcap=true - Run
pm_troubleshoot_quick_scanon the returnedlocal_path - Drill down with
pm_troubleshoot_flowon suspect IPs
Development
uv run pytest -q
uv run packetmaster-mcp # stdio MCP server
Security
- PCAP paths must be absolute; optional
PM_ALLOWED_DIRSsandbox for locked-down setups - Symlinks rejected
- Stream/field output redacts Authorization, Bearer tokens, passwords by default
- Confirm with user before
pm_follow_streamon production captures
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。