/ docs

MCP servers

Install and manage Model Context Protocol servers — 17-entry catalog + custom.

▶

MCP servers — walkthrough

Video walkthrough coming soon.

MCP servers — walkthrough

What is MCP

Model Context Protocol is a standard from Anthropic for letting LLM agents talk to external tools and data sources through a small JSON-RPC surface. An MCP server exposes tools + resources; the client (Hypex) connects to it, advertises the tools to the model, and proxies calls.

In practice: MCP = the plugin system for agentic IDEs. Filesystem, GitHub, Slack, Postgres, Puppeteer, Brave Search — each is one small server.

Installing from the marketplace

Hypex: MCP Marketplace opens a grid of 17 curated servers grouped by category (dev · docs · comms · data · web · productivity · search · utility). One-click install:

Pick a server
Fill in required env vars (e.g. GITHUB_PERSONAL_ACCESS_TOKEN) — Hypex prompts you via a password-masked input if the var is secret
Click install. The server config lands in .hypex/mcp.json
Hypex auto-starts it via npx -y <pkg> (or the command you specified)
Within a few seconds, the server's tools appear in the agent's tool list — auto-prefixed with the server id to avoid name clashes

Catalog highlights

Server	What it does
filesystem	Read / write / search files under a sandboxed root
github	Create issues / PRs, read repos, comment, search code
slack	Send / read channels, list users
postgres	Read-only SQL queries against a connection URL
sqlite	Read SQLite files
puppeteer	Headless browser — navigate, click, screenshot, scrape
brave-search	Web + news search through Brave's API
memory	Long-term key/value memory across sessions
fetch	HTTP GET / POST with response caching
git	Read git history + diffs without shelling out
time	Current time, timezone conversion
sequential-thinking	Step-by-step reasoning scratchpad
gdrive	List + read Google Drive files
notion	Read / write Notion pages + databases
sentry	Query errors, create issues
google-drive	Browse and read Drive
everything	Reference server with every MCP feature for testing

Config file

All servers persist to .hypex/mcp.json:

{
  "servers": {
    "filesystem": {
      "transport": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "${workspaceFolder}"]
    },
    "github": {
      "transport": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..." }
    }
  }
}

Edit directly with Hypex: Edit MCP Config. Save triggers an auto-reload.

Transports

stdio — most common, spawns the server as a subprocess with command + args, communicates over stdin/stdout
http — url pointing at an HTTP JSON-RPC endpoint
sse — server-sent events, url for the streaming channel

Managing servers

Hypex: MCP Marketplace — install from catalog
Hypex: Edit MCP Config — open .hypex/mcp.json
Hypex: Reload MCP Servers — re-read config + restart stdio servers
Hypex: Show MCP Logs — output channel with every RPC frame (useful when a server is misbehaving)
Hypex: Uninstall MCP Server — remove from config

Status bar

The left status bar shows an MCP pill when servers are active. Click to open the marketplace.

Security notes

MCP servers run as subprocesses on your machine with your user's permissions. Only install ones you trust — just like npm install.
Tool calls from MCP servers still go through the agent's approval model (write / exec tools prompt).
Env vars in mcp.json are stored in plaintext — OK for dev tokens, not OK for production secrets. Use hypex.mcp.secrets.* settings (coming) for keychain-backed values.

Writing your own server

Anthropic's MCP quickstart is the canonical reference. Hypex works with any spec-compliant 0.x server.