zephex
Zephex 是一个为 AI 编程编辑器构建的托管 MCP 网关。它为你的智能体提供 10 个即用工具——检查 npm 包漏洞、审计安全标头、读取和搜索代码、追踪请求流、从任何仓库获取项目上下文等。一个 API 密钥即可立即与 Claude Code、Cursor、VS Code、Windsurf 等工具配合使用。在 zephex.dev 免费开始使用。
文档
Works everywhere
One config. Every editor.
One endpoint for every code editor. Same tools, same config — 20+ editors supported.
Questions beforeyou connect
Quick answers for evaluation, setup, and security review.
Zephex is a hosted MCP gateway with ten codebase tools, plus a terminal CLI and browser web terminal on the same API key — so AI coding agents (and humans) work from real project context instead of guessing.
A few minutes. Create a free key, run npx zephex setup (or paste the MCP config), and call tools from Cursor, Claude Code, Codex, OpenCode, VS Code, or the CLI. No servers to host.
Hosted MCP works with Cursor, Claude Code, Codex, OpenCode, VS Code, Windsurf, Zed, JetBrains, and other MCP clients — plus the terminal CLI and web terminal without any AI editor open.
The zephex terminal CLI installs via curl install.sh or npx zephex. From your project directory you run overview, deep --json, find, read, architecture, safe / check-package, check test, remember/recall, and more. Same backend tools as MCP; human-readable output or --json for agents and scripts.
A dashboard browser terminal at /dashboard/terminal that runs allowlisted Mode 2 commands over SSE — paste a public GitHub repo and run overview, find, deep, package checks without local install. Not a full shell. Shares your plan quota. Private repos: use local CLI or editor MCP.
No. Zephex does not use your prompts, tool inputs, tool outputs, or code to train AI models. Security and data-use pages spell out boundaries.
A local MCP server is tied to one machine and one editor. You re-install it for every project, teammate, and laptop. Zephex is hosted — one endpoint, one key, same tools in every editor plus CLI and web terminal.
Context7 injects fresh third-party library docs. Zephex reads your actual repo and packages — project context, code search, architecture, tests, supply-chain checks. Recommended stack: Context7 + GitHub MCP + Zephex. See /compare/recommended-mcp-stack.
No. Mem0-style products store chat memory. Zephex is live engineering context (files, deps, architecture, package risk). project_memory stores project decisions and gotchas you choose to remember — not contacts or long chat logs.
Yes. check_package (CLI: zephex safe / check-package) hits live registry data for typosquats, maintainer signals, CVEs, and upgrade breakage — including the workflow that flagged a convincing fake Stripe SDK.
555 requests per month on Free (1 tool call ≈ 1 request), across MCP, CLI hosted tools, and web terminal. Pro is 3,500 at $7/mo; Max is 10,000 at $19/mo. See Pricing and /docs/plans.