io.github.danmartuszewski/hop
编码与调试by danmartuszewski
带有 MCP server 的 SSH 连接管理器,支持 AI 辅助在多台服务器间执行运维操作。
什么是 io.github.danmartuszewski/hop?
带有 MCP server 的 SSH 连接管理器,支持 AI 辅助在多台服务器间执行运维操作。
README
Why hop?
# Before: remembering and typing this every time
ssh -i ~/.ssh/work_key deploy@app-server-prod-03.us-east-1.example.com -p 2222
# After
hop prod
hop prod # fuzzy match any server
hop exec production "uptime" # run command on all prod servers
hop import # import your existing ~/.ssh/config
hop # launch the TUI, manage everything
Install
Homebrew (macOS/Linux)
brew install danmartuszewski/tap/hop
Go
go install github.com/danmartuszewski/hop/cmd/hop@latest
From source
git clone https://github.com/danmartuszewski/hop.git && cd hop && make build
./bin/hop
Install with an AI agent
Using Claude Code, Codex, Cursor, or another coding agent? Paste the block below into your agent and it will pick the right install path for your machine, register hop's MCP server, and verify the install.
Install hop on this machine and register its MCP server. Do the steps in order;
stop and report on the first failure.
1. Pick ONE install method, in this priority:
a. Homebrew (macOS or Linux):
brew install danmartuszewski/tap/hop
b. Go 1.22+ available:
go install github.com/danmartuszewski/hop/cmd/hop@latest
c. From source (no brew, no Go on PATH):
git clone https://github.com/danmartuszewski/hop.git && cd hop && make install
2. Verify the binary is on PATH:
hop version
3. Register the MCP server with whichever agent the user is running. Skip
clients the user does not use:
- Claude Code: claude mcp add hop -- hop mcp
- Codex CLI: codex mcp add hop -- hop mcp
- Claude Desktop / Cursor / Windsurf — add to the client's MCP config:
{ "hop": { "command": "hop", "args": ["mcp"] } }
4. (Optional) Seed the config from the user's existing SSH config. Preview
first; --yes is required for a non-interactive run:
hop import --dry-run
hop import --yes
5. Confirm hop's MCP tools are reachable from the agent (e.g. list_connections).
Constraints:
- Do NOT run bare `hop` — it launches an interactive TUI and will hang a
non-interactive session. Use subcommands (`hop version`, `hop list`, …).
- Do NOT modify ~/.ssh/config. hop reads it via `hop import` only.
- Do NOT commit secrets or identity files.
After step 3, restart the agent so it picks up the new MCP server.
Features
- Fuzzy matching - Type
hop prodto connect toapp-server-prod-03 - TUI dashboard - Browse, add, edit, delete connections with keyboard or mouse
- SSH config import - Already have servers in
~/.ssh/config? Import them in one command - Export - Export filtered connections to YAML for sharing or backup
- Multi-exec - Run commands across multiple servers at once
- Groups & tags - Organize by project, environment, or custom tags
- Jump hosts - ProxyJump support for bastion servers
- Landing directory - Drop straight into a predefined working directory on connect
- MCP server - Let AI assistants manage your servers — search connections, run commands, check status across projects
- Mosh support - Use mosh instead of SSH for roaming and unreliable connections
- Zero dependencies - Single binary, works anywhere
See all features in action: Demo recordings
Raycast Extension
Launch connections directly from Raycast. Fuzzy search, tags, environments - all at your fingertips.
<p align="center"> <img src="assets/hop1.png" width="32%"> <img src="assets/hop2.png" width="32%"> <img src="assets/hop3.png" width="32%"> </p>Configuration
Config file location: ~/.config/hop/config.yaml
version: 1
defaults:
user: admin
port: 22
# use_mosh: true # Uncomment to use mosh for all connections
connections:
- id: prod-web
host: web.example.com
user: deploy
identity_file: ~/.ssh/work_key # Private key for this connection
remote_dir: /var/www/myapp # Land in this directory on connect
project: myapp
env: production
tags: [web, prod]
- id: prod-db
host: db.example.com
user: dbadmin
port: 5432
project: myapp
env: production
tags: [database, prod]
- id: staging
host: staging.example.com
user: deploy
project: myapp
env: staging
- id: private-server
host: 10.0.1.50
user: admin
proxy_jump: bastion # Connect via jump host
forward_agent: true # Forward SSH agent
- id: remote-dev
host: dev.example.com
user: dan
use_mosh: true # Use mosh instead of SSH
groups:
production: [prod-web, prod-db]
web-servers: [prod-web, staging]
Security note:
forward_agent: trueexposes your SSH keys to anyone with root access on the remote server. Only enable this for servers you fully trust. Consider usingproxy_jumpinstead when you just need to reach internal hosts through a bastion.
Mosh Support
Mosh (mobile shell) is useful for connections over unreliable networks — it handles roaming, intermittent connectivity, and high latency gracefully.
Global default — enable mosh for all connections:
defaults:
use_mosh: true
connections:
- id: remote-dev
host: dev.example.com
- id: legacy-server
host: old.example.com
use_mosh: false # Override: use SSH for this one
Per-connection — enable mosh for specific connections:
connections:
- id: remote-dev
host: dev.example.com
user: dan
use_mosh: true
One-off — use the --mosh flag without changing config:
hop connect myserver --mosh
hop myserver --mosh
Per-connection use_mosh: false overrides the global default. SSH options (port, identity file, proxy jump, agent forwarding) are automatically passed to mosh via its --ssh flag. Mosh requires both the local mosh-client and mosh-server on the remote host.
Note:
hop execalways uses SSH regardless ofuse_mosh, since mosh is designed for interactive sessions.
Landing Directory
Set remote_dir to have a connection drop you straight into a specific directory instead of $HOME:
connections:
- id: prod-web
host: web.example.com
user: deploy
remote_dir: /var/www/myapp # cd here on connect
- id: my-dev
host: dev.example.com
remote_dir: ~/projects/api # ~ is expanded on the remote host
On connect, hop runs cd into the directory and then hands you a normal interactive login shell, so the session behaves exactly like a regular SSH login — just somewhere else. A few details worth knowing:
- Absolute paths and
~both work.~and~userare expanded by the remote shell. - Forgiving by design. If the directory is missing or inaccessible, you still land in a shell (in
$HOME) rather than getting bounced off the host. - Works in new tabs too.
hop opencarries the landing directory into every terminal it launches.
Note:
remote_diris ignored when you pass an explicit command (e.g.hop connect web -- uptimeorhop exec), since those aren't interactive sessions.
TUI Dashboard
Launch with hop or hop dashboard.
When you connect to a server from the dashboard (by pressing Enter), the SSH session starts, and the dashboard automatically returns after the session ends. This lets you quickly hop between servers without restarting the TUI each time.
For one-shot connections that exit to your terminal, use:
hop <query> # fuzzy match and connect
hop connect <id> # connect by exact ID
Keyboard Shortcuts
| Key | Action |
|---|---|
↑/k | Move up |
↓/j | Move down |
PgUp/PgDn | Move by page |
g | Go to top |
G | Go to bottom |
/ | Filter connections (supports multi-keyword AND search) |
t | Filter by tags |
r | Toggle sort by recent |
Enter | Connect to selected |
a | Add new connection |
i | Import from SSH config |
p | Paste SSH string (quick add) |
e | Edit selected |
c | Duplicate selected (opens a prefilled copy) |
d | Delete selected |
x | Export connections to YAML |
y | Copy SSH command |
T | Open theme picker |
? | Show help |
q | Quit |
Filtering Connections
Press / to filter connections by typing keywords. The filter supports multi-keyword AND logic - separate keywords with spaces to find connections matching all terms.
Examples:
prod- matches connections containing "prod"prod web- matches connections containing both "production" AND "web"kaf staging- matches connections with both "kafka" AND "staging"
The filter searches across connection IDs, hosts, projects, environments, and tags.
Quick Add with Paste
Press p and paste any of these formats:
user@host.com
user@host.com:2222
ssh user@host.com -p 2222
ssh://user@host:port
The connection form opens with fields pre-filled.
Duplicating a Connection
Press c on any connection to create a copy. The add form opens with every
field pre-filled from the original — including options that aren't shown in
the form, like proxy jump and mosh — and a collision-free ID suggestion (e.g.
web-prod → web-prod-copy). Adjust whatever you need and save. If you pick an
ID that already exists, the form stays open with your edits intact so you can
fix it.
Importing from SSH Config
Import existing connections from your ~/.ssh/config file:
From the dashboard: Press i to open the import modal, select which connections to import, and press Enter.
From the CLI:
hop import # Import from ~/.ssh/config
hop import --dry-run # Preview what would be imported
hop import --file ~/.ssh/config.d/work # Import from custom path
What gets imported:
- Host alias becomes the connection ID
- HostName, User, Port, IdentityFile
- ProxyJump for jump host connections
- ForwardAgent setting
What gets skipped:
- Wildcard patterns (
Host *,Host *.example.com) - Entries without a HostName (alias is used as hostname)
Conflict handling: If a connection ID already exists, the imported connection is renamed with -imported suffix (e.g., myserver → myserver-imported).
Exporting Connections
Export a subset of connections to a YAML file for sharing, backup, or transferring to another machine.
From the dashboard: Press x to open the export modal. Only currently filtered connections are shown — apply text or tag filters first to narrow the selection. Toggle items with Space, then press Enter to save.
From the CLI:
hop export --all # Export all to stdout
hop export --all -o backup.yaml # Export all to a file
hop export --project myapp -o myapp.yaml # Export by project
hop export --tag database # Export by tag
hop export --env production # Export by environment
hop export --id web-1,web-2 # Export specific connections
At least one filter flag or --all is required. Filters combine with AND logic.
Theming
The dashboard ships with sixteen color presets — each popular theme has both a dark and a light variant, listed separately so you can pick whichever you want regardless of your terminal background. Press T to browse them with live preview: ↑/↓ to navigate, Enter to save the choice into your config, Esc to revert.
| Family | Dark | Light |
|---|---|---|
| Built-in hop | default-dark | default-light |
| Everforest | everforest-dark | everforest-light |
| Gruvbox | gruvbox-dark | gruvbox-light |
| Catppuccin | catppuccin-mocha | catppuccin-latte |
| Tokyo Night | tokyo-night-storm | tokyo-night-day |
| Solarized | solarized-dark | solarized-light |
| Nord | nord | nord-light |
| Dracula | dracula | alucard |
Picking a preset writes a single line to your config:
theme_preset: everforest-dark
When theme_preset is unset, hop auto-picks default-dark or default-light based on your terminal background.
Custom overrides
Layer your own colors on top of any preset:
theme_preset: everforest-dark # optional; omit to auto-pick default
theme: # applies to every preset
primary: "#0066cc"
theme_dark: # only applies when the preset is a dark variant
selection: "#1f1f28"
theme_light: # only applies when the preset is a light variant
foreground: "#1c1f24"
Color values can be either a quoted ANSI 256 code ("39") or a hex string ("#bd93f9"). ANSI codes adapt to your terminal's palette; hex values are absolute.
Available keys: primary, secondary, accent, success, warning, error, muted, selection, foreground. Any key you don't set falls through to the preset, then to the built-in default.
CLI Commands
hop # Open TUI dashboard
hop <query> # Fuzzy match and connect
hop connect <id> # Connect by exact ID
hop get <id> <field> # Print single field value to stdout
hop get <id> f1,f2,f3 # Print multiple fields tab-separated
hop get <id> # Print all fields as "key value" lines
hop get --help # Full field list and flags
hop list # List all connections
hop list --json # List as JSON
hop list --flat # Flat list without grouping
hop import # Import from ~/.ssh/config
hop import --file <path> # Import from custom path
hop import --dry-run # Preview without importing
hop export --all # Export all connections to stdout
hop export --project <name> # Export filtered connections
hop export --tag <tag> -o f # Export to file
hop open <target...> # Open multiple terminal tabs
hop exec <target> "cmd" # Execute command on multiple servers
hop resolve <target> # Test which connections a target matches
hop mcp # Start MCP server (read-only)
hop mcp --allow-exec # Start MCP server with remote exec
hop version # Show version
Targeting
Commands like exec and open accept a target that resolves to one or more connections. The target is matched in this order:
- Named group — an explicit list of connection IDs defined under
groups:in config - Project-env pattern — matches connections by
projectandenvfields (e.g.myapp-prodmatches all connections withproject: myappandenv: prod) - Glob pattern — wildcard matching on connection IDs (e.g.
web*,*-prod-*) - Fuzzy match — falls back to fuzzy matching a single connection ID
You can also filter any target by tag with --tag.
Use hop resolve to preview which connections a target will match before running anything:
hop resolve production # see what "production" resolves to
hop resolve "web*" # test a glob pattern
hop resolve myapp-prod --tag=web # combine target + tag filter
Examples
# Fuzzy connect
hop prod # matches "prod-web", "prod-db", etc.
hop web # matches first *web* server
# Multi-exec with different target types
hop exec production "uptime" # named group
hop exec myapp-prod "df -h" # project-env pattern
hop exec "web*" "systemctl status" # glob pattern
hop exec --tag=database "psql -c '\\l'" # tag filter
# Open multiple tabs
hop open production # named group
hop open web1 db1 api1 # specific IDs
hop open myapp-prod -- "htop" # with initial command
# List connections
hop list --flat
Scripting with hop
hop get prints connection fields to stdout so you can drop them straight into shell pipelines and command substitutions — think of it as ssh -G for your hop config.
# Build an ssh invocation from config:
ssh -i "$(hop get prod identity_file)" "$(hop get prod user)@$(hop get prod host)"
# Read multiple fields at once (tab-separated):
IFS=$'\t' read -r host port user < <(hop get prod host,port,user)
# Fallback when a field is empty:
hop get prod port --default 22
# Strict shells: suppress the trailing newline.
hop get prod host -n
# Dump all non-empty scalar fields (ssh -G style "key value" lines):
hop get prod
# Read a single SSH option by key:
hop get prod options.StrictHostKeyChecking
# Structured output for jq and friends:
hop get prod host,port --json | jq -r .host
Matching is exact ID only (not fuzzy) — safer inside scripts. Unknown IDs exit 1 with a "did you mean" hint. See hop get --help for the full field list.
MCP Server (AI Assistant Integration)
hop includes a built-in Model Context Protocol server that lets AI assistants like Claude Code and Codex manage your servers directly. Ask your assistant to check disk space across production, restart a service on staging, or find which servers belong to a project — it discovers your connections, resolves targets, and executes commands through hop.
<p align="center"> <img src="assets/mcp.png" alt="Claude Code managing servers through hop's MCP server"> </p>Setup
Claude Code:
claude mcp add hop -- hop mcp
Codex CLI:
codex mcp add hop -- hop mcp
Claude Desktop — add to your config (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"hop": {
"command": "hop",
"args": ["mcp"]
}
}
}
Codex — add to ~/.codex/config.toml:
[mcp_servers.hop]
command = "hop"
args = ["mcp"]
Or generate the Claude Desktop config automatically:
hop mcp --print-client-config # read-only
hop mcp --print-client-config --allow-exec # with remote exec enabled
Tools
By default, only read-only tools are exposed:
| Tool | Description |
|---|---|
list_connections | List connections, filter by project/env/tag |
search_connections | Fuzzy search across all connections |
get_connection | Get details for a specific connection |
resolve_target | Preview how a target pattern resolves |
list_groups | List all named groups |
get_history | Connection usage history |
build_ssh_command | Build the full SSH command string |
To enable remote command execution, start with --allow-exec:
claude mcp add hop -- hop mcp --allow-exec
codex mcp add hop -- hop mcp --allow-exec
This adds the exec_command tool, which runs shell commands on matched servers with output limits (64KB/host, 50 hosts max).
Resources
The server also exposes browsable resources:
| URI | Description |
|---|---|
hop://config | Config summary (counts, projects, environments) |
hop://connections | All connections |
hop://connections/{id} | Individual connection details |
hop://groups | All groups and members |
Security
- Identity files (SSH key paths) are never exposed through MCP
- Remote execution is disabled by default and requires explicit
--allow-exec - All logging goes to stderr to keep the JSON-RPC transport clean
Shell Completions
# Bash (Linux)
hop completion bash | sudo tee /etc/bash_completion.d/hop > /dev/null
# Bash (macOS with Homebrew)
hop completion bash > $(brew --prefix)/etc/bash_completion.d/hop
# Zsh (add to ~/.zshrc)
source <(hop completion zsh)
# Fish
hop completion fish > ~/.config/fish/completions/hop.fish
Flags
-c, --config <path> # Use custom config file
-v, --verbose # Verbose output
-q, --quiet # Suppress non-essential output
--dry-run # Print SSH command without executing
--mosh # Use mosh instead of SSH for this connection
Building
make build # Build binary to ./bin/hop
make test # Run tests
make test-docker # Run tests in Docker (isolated)
make install # Install to $GOPATH/bin
make docker # Build Docker image
Docker
# Build image
docker build -t hop .
# Run interactively
docker run -it --rm hop
# Run tests in container
docker build --target tester -t hop-test .
Project Structure
hop/
├── cmd/hop/ # Main entry point
├── internal/
│ ├── cmd/ # CLI commands (cobra)
│ ├── config/ # Configuration loading/saving
│ ├── export/ # Export logic
│ ├── fuzzy/ # Fuzzy matching
│ ├── mcp/ # MCP server (tools, resources, types)
│ ├── picker/ # Connection picker (promptui)
│ ├── resolve/ # Target resolution logic
│ ├── ssh/ # SSH connection handling
│ ├── sshconfig/ # SSH config parsing
│ └── tui/ # TUI dashboard (bubbletea)
├── Dockerfile
├── Makefile
└── README.md
License
MIT License - see LICENSE for details.
常见问题
io.github.danmartuszewski/hop 是什么?
带有 MCP server 的 SSH 连接管理器,支持 AI 辅助在多台服务器间执行运维操作。
相关 Skills
前端设计
by anthropics
面向组件、页面、海报和 Web 应用开发,按鲜明视觉方向生成可直接落地的前端代码与高质感 UI,适合做 landing page、Dashboard 或美化现有界面,避开千篇一律的 AI 审美。
✎ 想把页面做得既能上线又有设计感,就用前端设计:组件到整站都能产出,难得的是能避开千篇一律的 AI 味。
网页应用测试
by anthropics
用 Playwright 为本地 Web 应用编写自动化测试,支持启动开发服务器、校验前端交互、排查 UI 异常、抓取截图与浏览器日志,适合调试动态页面和回归验证。
✎ 借助 Playwright 一站式验证本地 Web 应用前端功能,调 UI 时还能同步查看日志和截图,定位问题更快。
网页构建器
by anthropics
面向复杂 claude.ai HTML artifact 开发,快速初始化 React + Tailwind CSS + shadcn/ui 项目并打包为单文件 HTML,适合需要状态管理、路由或多组件交互的页面。
✎ 在 claude.ai 里做复杂网页 Artifact 很省心,多组件、状态和路由都能顺手搭起来,React、Tailwind 与 shadcn/ui 组合效率高、成品也更精致。
相关 MCP Server
GitHub
编辑精选by GitHub
GitHub 是 MCP 官方参考服务器,让 Claude 直接读写你的代码仓库和 Issues。
✎ 这个参考服务器解决了开发者想让 AI 安全访问 GitHub 数据的问题,适合需要自动化代码审查或 Issue 管理的团队。但注意它只是参考实现,生产环境得自己加固安全。
Context7 文档查询
编辑精选by Context7
Context7 是实时拉取最新文档和代码示例的智能助手,让你告别过时资料。
✎ 它能解决开发者查找文档时信息滞后的问题,特别适合快速上手新库或跟进更新。不过,依赖外部源可能导致偶尔的数据延迟,建议结合官方文档使用。
by tldraw
tldraw 是让 AI 助手直接在无限画布上绘图和协作的 MCP 服务器。
✎ 这解决了 AI 只能输出文本、无法视觉化协作的痛点——想象让 Claude 帮你画流程图或白板讨论。最适合需要快速原型设计或头脑风暴的开发者。不过,目前它只是个基础连接器,你得自己搭建画布应用才能发挥全部潜力。