reach

name	reach
description	Manage remote servers via Reach — AI-native alternative to SSH. Use when user asks to check servers, deploy, run remote commands, read/write remote files, or monitor system health.

Reach — Remote Server Management

You have access to remote servers via Reach MCP tools. Reach replaces SSH with HTTPS + Token — you get structured JSON, not text to parse.

Setup

If reach tools aren't available, tell the user:

reach mcp install --global  # then restart your AI client

Tools

Tool	Use When
reach_dryrun	ALWAYS before destructive commands (rm, mv, kill, restart)
reach_stats	Need CPU/memory/disk/network stats — NOT reach_bash + top/free
reach_read	Read remote files — NOT reach_bash + cat
reach_write	Write remote files (atomic) — NOT reach_bash + echo >
reach_upload	Send local file to server
reach_bash	Run any shell command (fallback when no specialized tool fits)
reach_list	Discover available servers
reach_info	Quick server identity (hostname, OS, arch)

Rules

1. Dryrun before destroy. Before any reach_bash that modifies state, call reach_dryrun first. If risk is high or blocked, warn the user.

2. Structured over shell. Use reach_stats instead of parsing top/free/df output. Use reach_read/reach_write instead of cat/echo.

3. List first. If you don't know server names, call reach_list.

Patterns

Safe config change:

old = reach_read(server, path)        # backup
reach_write(server, path, new_config) # atomic write
result = reach_bash(server, "nginx -t") # validate
if failed: reach_write(server, path, old) # rollback
else: reach_bash(server, "systemctl reload nginx")

Health check:

stats = reach_stats(server)
# stats.cpu.usage_percent, stats.memory.usage_percent, stats.disk.partitions[].usage_percent
# All numbers, no parsing needed

Before dangerous command:

check = reach_dryrun(server, "rm -rf /opt/old")
# check.risk = "high", check.score = 85, check.affected.files = 847
# Warn user, then proceed only if confirmed