remote-bridge-cli

v0.1.6

Published

4 days ago

Bridge the gap between AI-generated code and remote servers

0High
0Medium
0Low

saivaraprasadreddymedapati

ai ssh rsync mcp deployment remote claude gemini

RemoteBridge 🌉

Bridge the gap between AI-generated code and remote servers.

RemoteBridge is a high-performance CLI tool written in Rust that acts as a stateful proxy between your local AI coding environment (Claude Code, Gemini CLI, Aider, etc.) and your remote infrastructure. It parses Markdown output from AI tools, syncs file changes via rsync, executes remote commands over SSH, and pipes logs back to your terminal — so the AI can "see" remote errors and fix them.

💬 Just Talk To Your AI

Once installed, you don't run commands manually. You just tell your AI what you want — in plain English — and RemoteBridge handles it.

You:  "Sync my project files to the staging server"
You:  "Run npm install on the remote server"
You:  "Deploy my latest changes to [email protected]"
You:  "Check what OS and runtimes are installed on the server"
You:  "Tail the remote logs and show me what's failing"
You:  "Something broke after deploy — fetch the logs and fix it"
You:  "Push my code and restart the app"

Your AI calls the right tool, syncs the right files, runs the right commands — and if something fails, it reads the remote logs and fixes the code automatically.

No manual SSH. No piping. No copy-pasting errors.

🚀 The Problem & Solution

The Problem: AI coding tools operate locally. Developers without CI/CD must manually FTP files and SSH into servers to test changes. The AI never sees remote runtime errors, creating a "context gap."

The Solution: RemoteBridge automates the Sync → Run → Feedback loop. It parses AI output, applies file changes locally, syncs them to a remote target, runs commands, and tails remote logs if something fails.

✨ Features

🛠 Get Started in 3 Steps

Step 1 — Install

Prerequisites: Rust (rustup.rs), rsync, ssh in PATH

npm install -g remote-bridge-cli

Builds a native binary optimized for your machine.

Step 2 — Add to your AI IDE

# Claude Code (available in every project automatically)
claude mcp add remote-bridge --scope user -- remote-bridge mcp

For other tools see the MCP configuration section below.

Step 3 — Point it at your server

Run this once per project:

remote-bridge init --name my-app -H your-server.com --user ubuntu --path /var/www/app

That's it. Now open your AI and just talk to it.

Build from Source

git clone https://github.com/varaprasadreddy9676/remote-bridge.git
cd remote-bridge
cargo build --release
cp target/release/remote-bridge /usr/local/bin/

⚙️ Configuration (`remotebridge.yaml`)

Initialize a project:

remote-bridge init --name my-app -H 13.234.xx.xx --user ubuntu --path /var/www/html/app

This creates a remotebridge.yaml:

project_name: "my-app"
targets:
  staging:
    host: "13.234.xx.xx"
    user: "ubuntu"
    remote_path: "/var/www/html/app"
    port: 22                          # optional, default: 22
    ssh_key: "~/.ssh/id_rsa"          # optional, uses default SSH key if omitted
    restart_cmd: "pm2 restart app"    # optional, used by restart & deploy commands
    logs:
      - "/var/www/html/app/logs/error.log"
      - "/var/log/nginx/error.log"
    require_confirmation: false        # set true to confirm every command
    exclude:                           # extra rsync exclusions beyond .gitignore
      - "node_modules/"
      - "*.log"
    blocked_patterns:                  # always rejected — no AI override
      - "rm -rf"
      - "drop table"
    allowed_commands:                  # when set, only these prefixes can run
      - "npm"
      - "pm2"
    audit_log: "~/.remote-bridge-staging.log"  # default: ~/.remote-bridge-audit.log
  production:
    host: "prod.example.com"
    user: "ubuntu"
    remote_path: "/opt/app"
    ssh_key: "~/keys/prod.pem"
    restart_cmd: "systemctl restart myapp"
    logs:
      - "/opt/app/logs/error.log"
    require_confirmation: true

📖 Commands

Core Workflow

Service Management

Monitoring

Developer Loop

🤖 Integrating with AI CLIs

RemoteBridge is pipe-friendly. Pipe any AI tool's output directly into remote-bridge apply.

With Claude Code

claude "Fix the database connection in src/db.ts and restart the app" --non-interactive \
  | remote-bridge apply --target staging

With Gemini CLI

gemini "Add rate limiting to the Express API" \
  | remote-bridge apply --target staging

With Aider

aider --message "Refactor the login logic" --apply \
  | remote-bridge apply --target staging

The Full AI Loop (Recommended)

# 1. Start watching in one terminal — changes sync automatically
remote-bridge watch --target staging

# 2. Run AI in another terminal
claude "Optimize the database queries in src/db.ts"

# 3. Deploy when ready
remote-bridge deploy --target staging --follow

🔌 MCP Support — Works With Every AI IDE

RemoteBridge is a native Model Context Protocol (MCP) server. Any MCP-compatible AI IDE can use it as a tool directly — no piping required. The config is almost identical across all tools.

Available MCP Tools

Claude Desktop

File: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "remote-bridge": {
      "command": "remote-bridge",
      "args": ["mcp"]
    }
  }
}

Cursor

File: ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project)

{
  "mcpServers": {
    "remote-bridge": {
      "command": "remote-bridge",
      "args": ["mcp"]
    }
  }
}

Or go to Cursor Settings → MCP → Add Server.

VS Code (GitHub Copilot / Continue / Cline)

File: .vscode/mcp.json in your project root

{
  "servers": {
    "remote-bridge": {
      "type": "stdio",
      "command": "remote-bridge",
      "args": ["mcp"]
    }
  }
}

For the Cline extension, add via the Cline sidebar → MCP Servers → Configure.

Windsurf (Codeium)

File: ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "remote-bridge": {
      "command": "remote-bridge",
      "args": ["mcp"]
    }
  }
}

Or go to Windsurf Settings → Cascade → MCP Servers → Add.

Zed

File: ~/.config/zed/settings.json

{
  "context_servers": {
    "remote-bridge": {
      "command": {
        "path": "remote-bridge",
        "args": ["mcp"]
      }
    }
  }
}

OpenAI Codex CLI

File: ~/.codex/config.json

{
  "mcpServers": {
    "remote-bridge": {
      "command": "remote-bridge",
      "args": ["mcp"]
    }
  }
}

Or pass inline: codex --mcp-server "remote-bridge mcp" "Deploy my changes"

Continue.dev

File: ~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "remote-bridge",
      "command": "remote-bridge",
      "args": ["mcp"]
    }
  ]
}

Claude Code (CLI)

File: .claude/mcp.json in your project, or ~/.claude/mcp.json globally

{
  "mcpServers": {
    "remote-bridge": {
      "command": "remote-bridge",
      "args": ["mcp"]
    }
  }
}

Or add via CLI: claude mcp add remote-bridge -- remote-bridge mcp

Any MCP-compatible tool

The pattern is always the same — stdio transport, command remote-bridge, arg mcp:

{
  "command": "remote-bridge",
  "args": ["mcp"],
  "transport": "stdio"
}

What it actually looks like

Once configured, you just talk to your AI naturally inside Claude Code, Cursor, Windsurf, or any MCP-enabled IDE:

You: "Sync my project files to the staging server"

RemoteBridge calls sync_to_remote → rsync transfers only changed files

You: "Run npm install on the remote server"

RemoteBridge calls run_remote_command with npm install → streams output back

You: "Deploy my latest changes to [email protected]"

RemoteBridge calls deploy → syncs files, restarts service, tails logs if it fails

You: "Check what OS and runtimes are installed on the server"

RemoteBridge calls preflight_check → returns Ubuntu version, Node, Python, Docker

You: "Tail the remote logs and show me what's failing"

RemoteBridge calls fetch_logs → AI reads the error and fixes your code

You: "Something broke after deploy — fetch the logs and fix it"

RemoteBridge fetches logs → AI sees the stack trace → writes the fix → deploys again

The AI decides which tool to call. You just describe what you want.

🔒 SSH Authentication & Security

RemoteBridge uses your existing SSH key authentication — no passwords stored.

Setup

Copy your SSH key to the server:

ssh-copy-id -i ~/.ssh/id_rsa.pub [email protected]

Use a specific key or non-standard port in config:

targets:
  staging:
    host: "13.234.xx.xx"
    user: "ubuntu"
    port: 2222
    ssh_key: "~/keys/staging.pem"

Or use ~/.ssh/config for aliases:

Host staging-server
    HostName 13.234.xx.xx
    User ubuntu
    IdentityFile ~/keys/my-key.pem
    Port 22

Then in remotebridge.yaml:

host: "staging-server"
user: "ubuntu"

🛡 Safety Features

RemoteBridge has layered defenses so a hallucinating AI can never destroy your server.

Confirmation Gate (built-in)

Commands containing sudo, rm, drop, delete, database, shutdown, reboot, killall, curl | bash, wget | sh, and 20+ other dangerous patterns always pause for confirmation before running on the remote host.

Hard Block (user-defined)

Patterns you add to blocked_patterns are always rejected — no confirmation, no override:

targets:
  staging:
    blocked_patterns:
      - "rm -rf"
      - "drop table"
      - "truncate"

Allowlist (user-defined)

When allowed_commands is set, only commands matching those prefixes can run. Anything else is silently blocked:

targets:
  production:
    allowed_commands:
      - "npm"
      - "pm2"
      - "systemctl restart myapp"

Audit Log

Every command execution is logged automatically to ~/.remote-bridge-audit.log (or a custom path):

[1742660591] host=your-server.com path=/var/www/app exit=0 cmd=npm install
[1742660612] host=your-server.com path=/var/www/app exit=-2 cmd=rm -rf /var/data

Exit codes: 0+ = actual exit code, -1 = skipped by user, -2 = hard blocked, -3 = not in allowlist.

Custom path:

targets:
  staging:
    audit_log: "/var/log/remote-bridge-staging.log"

Other Protections

require_confirmation: true — confirm every command, no exceptions
--dry-run on sync — preview rsync diff without touching anything
No password storage — SSH key auth only
No secrets in config — host/user/path only
MCP output truncation — run_remote_command returns at most 100 lines by default, preventing log floods from filling AI context

🤝 Contributing

Contributions, issues, and feature requests are welcome at GitHub Issues.

📜 License

MIT License.