ProxyKit-mcp
MCP server for ProxyKit — drive a local HTTP(S) debugging proxy from Claude, Cursor, and any MCP host.
README
proxykit-mcp
Model Context Protocol server for ProxyKit — drive a local HTTP(S) debugging proxy from Claude, Cursor, and any MCP host.
proxykit-mcp is a small stdio binary that exposes a running ProxyKit engine to
MCP clients. Every tool call is one HTTP call to the engine's control API on
127.0.0.1 — the server holds no traffic of its own and talks to nothing off
your machine.
"Mock the checkout endpoint to fail half the time." · "Summarize the last 500 captured requests against my baseline." · "Add 800 ms of latency to /api/payments and replay the session."
Typed into your editor, run against real traffic ProxyKit already captured.
- Website: https://proxykit.net
- Setup guide (all hosts): https://proxykit.net/mcp
- CLI reference: https://proxykit.net/docs
This repository is the public home of the proxykit-mcp server — its docs,
tool reference, and MCP registry metadata. The engine and desktop app are
ProxyKit, a proprietary product with a free tier.
Install
The binary ships inside every ProxyKit CLI archive and inside the desktop app. Pick whichever you already have.
With Homebrew (CLI)
brew install getproxykit/tap/proxykit-cli
That puts proxykit, proxy-engine, and proxykit-mcp on your PATH.
With WinGet (Windows)
winget install ProxyKit.ProxyKitCLI
From the desktop app
Download from https://proxykit.net/download. The binary is bundled at:
macOS /Applications/ProxyKit.app/Contents/Resources/proxykit-mcp
Windows C:\Program Files\ProxyKit\resources\proxykit-mcp.exe
Linux (AppImage) squashfs-root/resources/proxykit-mcp
Connect an MCP host
The engine writes a control.token file on startup; point the MCP server at
that file — never paste the token into the config. Add this to your host's
MCP config (~/.cursor/mcp.json, Claude Desktop's claude_desktop_config.json,
.mcp.json, …):
{
"mcpServers": {
"proxykit": {
"command": "proxykit-mcp",
"env": {
"PROXYKIT_CONTROL_AUTH_TOKEN_FILE": "~/.config/ProxyKit/control.token",
"PROXYKIT_MCP_TIER": "readonly"
}
}
}
}
Start ProxyKit (desktop or proxykit start --headless), restart your MCP host,
and ask it: "List my latest captured traffic in ProxyKit."
Per-client paths and one-click install from the desktop app are documented at https://proxykit.net/mcp.
Environment variables
| Variable | Purpose |
|---|---|
PROXYKIT_CONTROL_AUTH_TOKEN_FILE |
Path to the engine's control.token. Preferred over the raw token. |
PROXYKIT_CONTROL_AUTH_TOKEN |
The token directly (use the _FILE form instead where possible). |
PROXYKIT_MCP_TIER |
Comma-separated allowlist: readonly (default), mutate, capture, analysis, replay. |
PROXYKIT_ENGINE_URL |
Engine control API. Defaults to http://127.0.0.1:17171. |
Safety model
- Local only. The server calls
127.0.0.1:17171. The engine binds loopback and rejects non-loopback callers. - Token-gated. Every control call carries
X-Proxykit-Control-Token. No token, no access. - Least privilege by tier.
readonlyis the default and can only read. Anything that mutates rules, controls capture, or replays traffic is off until you opt its tier in — so an agent can't create a mock or clear your capture unless you explicitly allowed that tier. - Redacted at the source. Response bodies are redacted server-side before they reach the agent.
Tools
50 tools across five tiers. readonly is enabled by default; the rest are
opt-in via PROXYKIT_MCP_TIER.
readonly — read state (default)
| Tool | Does |
|---|---|
proxy_status |
Is the proxy running, what port, TLS configured, buffer size |
list_traffic |
List captured requests; filter by method/URL/status, paged |
get_request |
Full headers + redacted body of one request |
search_traffic |
Substring search across captured URLs |
list_mock_rules |
Configured mock rules |
list_rewrite_rules |
Configured rewrite rules |
list_chaos_rules |
Configured chaos (latency/error injection) rules |
list_scripts |
Installed JS hook scripts |
list_sessions |
Named capture sessions |
get_session |
One session's manifest + baseline |
list_findings |
Privacy / schema / performance / lint findings |
get_findings_summary |
Finding counts by severity and category |
list_environments |
Environments and the active one (secrets redacted) |
list_paused_breakpoints |
Requests paused awaiting review |
list_rule_packs |
Installed rule packs |
get_ca_status |
Whether a CA root is available (public PEM only) |
mutate — edit rules
create_mock_rule, update_mock_rule, delete_mock_rule, toggle_mock_rule,
create_mock_from_request, create_rewrite_rule, update_rewrite_rule,
delete_rewrite_rule, toggle_rewrite_rule, create_chaos_rule,
update_chaos_rule, delete_chaos_rule, toggle_chaos_rule,
attach_environment
capture — control capture
start_capture, stop_capture, clear_traffic, release_breakpoint
analysis — AI + findings workflows
analyze_request, rescan_findings, dismiss_finding, restore_finding,
propose_redaction_rules, run_ai_workflow, compare_sessions
replay — sessions, replay, rule packs
start_session, stop_session, delete_session, replay_session,
replay_request, import_rule_pack, install_rule_pack, export_rule_pack,
delete_rule_pack
License
This repository — the docs, Dockerfile, and registry metadata — is MIT
licensed (see LICENSE); reuse it freely.
ProxyKit itself — the proxykit-mcp binary and the engine it drives — is
proprietary software © 2026 Seyed Ahmad Sarollahi, licensed under
https://proxykit.net/terms#license. Nothing in this repository grants any right
to the ProxyKit software; the binary is downloaded from the release server at
build time under its own terms.
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。