Wireshark-MCP is an MCP (Model Context Protocol) server designed to provide AI assistants with packet analysis capabilities. It wraps tshark (and optionally Wireshark suite tools), turning them into a structured analysis interface. Users only need to hand a .pcap file to an AI client and ask questions in natural language to get answers based on real tshark output.
Main capabilities:
- Provides 52 tools, each based on real tshark output and organized by category.
- A single tool can cover 20 protocols, for example wireshark_analyze_protocol accepts protocol parameters (such as tls_handshakes, mqtt, modbus, s7comm, zigbee, wifi, rtp, kerberos, etc.) and automatically applies the correct fields and display filters.
- At server startup, only tshark is required; optional tools (capinfos, mergecap, editcap, dumpcap, text2pcap) are automatically detected and enable additional features when present.
- Tool results have a length limit: output exceeding 8000 characters is truncated at the head and tail and marked, and the remaining content can be viewed in pages via the tool's offset / limit / display_filter parameters.
- Each tool declares whether it is a read or write operation; clients can automatically approve the 41 read-only analysis tools, while still prompting for the 11 file-creating tools (live capture, merge, filter and save, editcap, text2pcap, frame extraction, object export).
- In version 3.0, these 11 write tools fail closed until WIRESHARK_MCP_ALLOWED_DIRS specifies an existing directory. Remote HTTP/SSE binding also remains loopback-only unless --allow-insecure-http is explicitly provided behind a trusted authenticated TLS proxy.
Usage:
- Prerequisites: Python 3.10+ and Wireshark's tshark in PATH.
- Wireshark MCP 3.0 uses the stable MCP Python SDK 2.x series (mcp>=2.1.1,<3).
- After installation, restart the AI client.
- If problems occur, run wireshark-mcp doctor for diagnostics. For manual configuration or platform-specific instructions, see docs/manual-configuration.md.
- Point the AI client at a .pcap file and ask questions in natural language.
Use cases:
- Network packet analysis, protocol statistics, field extraction, and security analysis.
- Works with 18+ MCP clients such as Claude Desktop, Claude Code, Cursor, VS Code, and more.
Limitations:
- Requires Python 3.10+ and tshark.
- The tool list is transmitted with the prompt prefix of every request, about 22 KB by default, which is a fixed per-request cost.
- The number of advertised tools can be reduced via --profile, suitable for clients that never capture live traffic or write pcap.
- Write tools fail closed by default and require configuring allowed directories.
- Remote HTTP/SSE binding is loopback-only by default and requires explicitly allowing insecure HTTP and placing it behind a trusted authenticated TLS proxy.
