mitmAny
Supports Android USB capture workflows, allowing an Android device with USB debugging enabled to route traffic through mitmAny for HTTP/HTTPS inspection without shared Wi-Fi.
Give your AI eyes on real network traffic. Turn captured requests into an API investigation.
mitmAny is an AI-ready HTTP / HTTPS traffic inspector for macOS, Windows and Android USB workflows. Capture what a website or app actually sends, connect Codex or another compatible MCP client, and ask AI to help reverse-engineer APIs from real evidence: endpoints, authentication, pagination, response schemas and business flows.
100% free and open source. Every mitmAny feature is free: no paid edition, subscription, trial expiry or feature paywall. Optional tips fund maintenance and unlock nothing. Any fees charged by your chosen AI client or model provider are separate from mitmAny.
Downloads
v1.2.1 · Release notes
Platform | Direct download |
macOS — Apple Silicon and Intel | |
Windows — x64 |
SHA-256 checksums. Extract before launching; move the macOS app to Applications. Windows requires WebView2. These packages are unsigned; macOS is not notarized. See installation and verification.
Related MCP server: mcp-network-analyzer
Inside mitmAny

This screenshot uses synthetic data. See the demo walkthrough.
From captured traffic to AI-assisted reverse engineering
flowchart LR
A["Operate an app"] --> B["Capture traffic"]
B --> C["Analyze with AI"]
C --> D["API evidence"]You operate the app. mitmAny captures the evidence. Your AI client helps make sense of it.
What you want to understand | What you can ask AI to investigate |
An undocumented API | Group observed endpoints, methods and parameters into an API outline, citing request IDs. |
Login and session behavior | Trace observed headers, cookies and request order; distinguish evidence from assumptions. |
Pagination and search | Compare requests while you scroll, change filters or search; identify cursor and page parameters. |
A business workflow | Mark actions and inspect the corresponding time windows to map a sequence of requests. |
A bug or unexpected response | Compare successful and failed requests, status codes, timing and captured bodies. |
Integration documentation | Draft schema notes and client-code examples from captured evidence for you to review. |
MCP is the bridge between your capture session and your AI tools. Instead of repeatedly pasting screenshots and JSON, the AI client can inspect capture status, scope an observation to a host, summarize/search flows, read headers and bounded bodies, and wait for new traffic. Observation bookmarks help organize an investigation; they do not start capture or preserve evicted records.
The bundled mitmany-network-analysis skill guides evidence-based analysis. Settings → Codex installs the skill and registers the MCP connection locally. Reconnect MCP or open a new Codex session, then start capture yourself and try:
Analyze the login and pagination protocol of
api.example.test. I will use the app. Scope the observation to this host, cite request IDs for your conclusions, and call out missing or truncated evidence. Finish with an API outline and a list of uncertainties.
AI analysis runs in your connected client, not in an embedded model inside mitmAny. MCP reads captured traffic and manages observation bookmarks; it does not replay requests, change proxies or bypass certificate pinning. Use it for systems you own or are authorized to inspect. Connection details (Chinese).
Language
English is the default language. Switch to 简体中文 in Settings → Language; your choice is saved for future launches. README translations are available in English, Simplified Chinese, Traditional Chinese, Japanese and Korean. The application interface supports English and Simplified Chinese only.
Get started
Download a package from Releases. Move the macOS app to Applications; Windows requires WebView2.
Open Settings → CA Certificate, then install and trust your local CA to inspect HTTPS. Without trust, HTTPS is tunneled without decryption.
Start desktop capture, or connect an Android device with USB debugging enabled and turn on its proxy in the Android panel. Use your browser or app, then select a request.
Stop capture when finished. Restore the Android proxy before unplugging USB. Use the toolbar recovery button if desktop networking needs restoration. See recovery and removal if the app cannot start.
Features
HTTP / HTTPS, HTTP/2, streaming responses and SSE prefixes.
Domain grouping, filters, search, favorites, virtual lists, JSON folding and body search.
Android USB capture without shared Wi-Fi.
HAR import/export, copy as cURL, captured-body export.
Local CA management, proxy backup/recovery, light and dark themes.
Read-only MCP tools and a bundled protocol-analysis skill; connect through Settings → Codex.
Limits and privacy
No TUN, UDP, QUIC or HTTP3 interception, certificate-pinning bypass, request replay or WebSocket message inspector. Clients must honor the proxy and trust the CA; Android apps may need explicit debug trust configuration.
Request and ordinary response capture: 1 MiB each. HTML response capture: 8 MiB. Decompression has the same limits. Retention: 3,000 requests and 256 MiB of body files. Busy queues may skip or truncate captures while forwarding continues. Observations do not pin evidence.
Traffic and CA keys stay local to mitmAny. Connected AI clients may send tool results to their model provider. Redaction covers common sensitive fields, not all possible secrets. HAR, cURL and body exports may contain credentials. Inspect only traffic you are authorized to access.
Privacy · Security reporting · Detailed Chinese manual
Development
Go 1.26+, Node.js from .nvmrc, Wails CLI 2.15.0 and native platform build tools are required.
nvm use
npm --prefix frontend ci
npm --prefix frontend test
npm --prefix frontend run build
go install github.com/wailsapp/wails/v2/cmd/wails@v2.15.0
wails devDevelopment · Architecture · Roadmap · Changelog · Releases
Support the project
If mitmAny is useful to you, you can leave an optional, one-time tip to support ongoing maintenance:
These are one-time payments, not subscriptions. Support does not unlock paid features or priority support. The Stripe checkout merchant is SimiSF.
Prices are set in USD and include applicable taxes. Stripe may offer a local-currency amount; review the currency and total before paying. Available payment methods depend on your region and checkout currency. Alipay and WeChat Pay were available in the CNY checkout when verified; WeChat Pay was not available in USD. The options shown at checkout take precedence.
Bug reports, documentation improvements and code contributions are also welcome.
License
MIT. Vendored dependencies retain their own licenses. See third-party notices.
Available Tools
9 toolsbegin_observationA
Bookmark a new observation starting now, optionally scoped by exact host. Does not start capture or clear traffic. Returns id and start_cursor.
Inputs: host.
capture_statusARead-only
Read capture state, cursor, data quality and retention limits. No input needed.
Inputs: no named parameters.
end_observationA
Close an observation time window, retaining bookmarks. Requires observation_id. Does not stop proxy capture.
Inputs: observation_id (required).
get_flowARead-only
Read headers, timing, capture counts and errors without body I/O. Requires flow_id. Optional observation_id enforces scope. Known sensitive headers and query fields are redacted.
Inputs: flow_id (required), observation_id.
mark_observationB
Add a label to an observation at the current time. Requires observation_id and label. Marks are temporal evidence, not proof of causality.
Inputs: label (required), observation_id (required).
read_bodyARead-only
Read bounded decoded text for flow_id and direction (request or response). offset and limit count Unicode code points after redaction; default 8000/max 32000. Decoder reads at most 1 MiB even for HTML. Known JSON/form secret keys redacted; arbitrary text is not guaranteed secret-free. Read completed flows for stable pagination.
Inputs: flow_id (required), direction (required), limit, offset, +1 more.
search_flowsARead-only
Search retained flow metadata by observation_id, exact host, path substring, method and status. Newest revision first. limit default 30 max 100. For pagination retain after_cursor and set before_cursor to next_before_cursor. reset indicates cursor history loss; snapshot fallback may contain older flows.
Inputs: host, path, limit, method, +4 more.
summarize_trafficARead-only
Group retained flows by exact host, method, path and status. Optional observation_id, host, path substring, method, status, limit (default 30, max 100). Returns counts and example IDs; paths are not inferred route templates.
Inputs: host, path, limit, method, +2 more.
wait_for_flowsARead-only
Wait for new or updated matching flows after required after_cursor. timeout_ms 0..30000 (default immediate). Optional observation_id and search filters. Returns on matching flows, timeout, cursor reset or cancellation. Follow pagination before advancing cursor.
Inputs: after_cursor (required), host, path, limit, +5 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 9 tools
Most tools have clearly distinct roles: observation lifecycle (begin/end/mark), metadata access (get_flow), body access (read_body), and status (capture_status). The one soft overlap is summarize_traffic versus search_flows, since both filter on observation_id, host, path, method, and status; the descriptions differentiate them (aggregated counts versus newest-first ID listing with cursors), but an agent could still pick the wrong one for a quick lookup.
Eight of nine names follow a consistent snake_case verb_noun pattern (begin_observation, end_observation, get_flow, read_body, search_flows, wait_for_flows, summarize_traffic, mark_observation). capture_status is the lone noun-only outlier, a minor deviation that stays readable.
Nine tools is well within the ideal 3-15 range and each earns its place: two observation boundary operations, one annotation, one status read, and five flow inspection/streaming operations. Nothing feels redundant or padded.
The surface covers the core lifecycle: create/end/mark observations, search and summarize flows, read metadata and bounded bodies, and wait for new traffic. The main gap is the absence of a list_observations operation, so an agent that loses an observation_id has no way to rediscover bookmarks; capture start/stop and traffic clearing appear intentionally out of scope but are noted as external.
Maintenance
Related MCP Connectors
Capture, inspect & debug HTTPS traffic across iOS, Android, browsers & backends — 304 MCP tools.
Find, document and execute authenticated API calls across 1000+ apps via one MCP server
Free MCP debugging tools for AI agents and developers. Diagnose failed API and MCP tool calls, identify HTTP errors, validate tool arguments against supported JSON Schema rules, and get actionable troubleshooting recommendations.
2,000+ MCP servers read at source level. Know what one does before you connect. Free, no key.
Related MCP Servers
- AlicenseAqualityDmaintenanceAn MCP server that enables AI assistants to capture and analyze HTTP/HTTPS traffic from Android devices. It supports smart searching of network requests and provides tools for detailed traffic analysis via natural language.11245MIT
- AlicenseNot gradedqualityCmaintenanceCaptures browser network traffic, analyzes API patterns, and exposes analysis tools through an MCP server for AI-assisted workflows.1MIT
- AlicenseAqualityDmaintenanceMCP server for intercepting and mocking HTTP(S) traffic via a Mockttp proxy, with tools for Android emulator setup, traffic inspection, protobuf analysis, and rule-based manipulation.35MIT
- FlicenseBqualityDmaintenanceA lightweight MCP server built on mitmproxy that lets LLMs capture, inspect, replay, and modify HTTP traffic through a set of focused tools.138-