Get Started
MCP is the protocol layer that lets AI coding tools move beyond the context window and call real tools over a consistent interface.
MCP is a standard JSON-RPC protocol. The same agent code talks to any MCP server. Zephex is one MCP server that wraps ten tools.
MCP, or Model Context Protocol, is an open standard developed by Anthropic that defines how AI models communicate with external tools and data sources. It gives AI coding agents a structured way to call tools, read files, search codebases, audit security, and perform other tasks through a consistent JSON-RPC 2.0 interface.
Before MCP, AI coding assistants could only work with what was already in their context window. With MCP, they can reach outside that window and interact with real systems: your code, your packages, your HTTP endpoints, and your documentation.
An MCP session usually works like this:
tools/list to discover what tools are available.tools/call with a tool name and arguments.The full exchange uses JSON-RPC 2.0, a simple request/response protocol carried over HTTP or stdio.
| Transport | How it works | Who manages the server |
|---|---|---|
| stdio | Server runs as a local subprocess and communicates over stdin/stdout. | You, locally on your machine. |
| HTTP (Streamable) | Server runs remotely and communicates over HTTPS. | The provider, in the cloud. |
Most open-source MCP setups use stdio. You run a local server command and your editor talks to it as a subprocess. That works, but it also means you manage packages, processes, and config on every machine where you want MCP available.
HTTP transport changes the model entirely. Instead of running servers locally, your editor connects to a remote HTTPS endpoint. That is what Zephex provides.
Zephex runs an MCP HTTP server at https://zephex.dev/mcp. Every request is authenticated with a Bearer token. The 10 available tools cover the most common AI coding agent needs: project orientation, code reading, pattern search, package intelligence, security auditing, and structured reasoning.
You configure your editor once with the endpoint URL and your API key. From that point on, every AI session in that editor has access to all 10 tools without any additional setup.
Start with the Quickstart to connect your first editor in under 5 minutes.