Servonaut MCP Server

Servonaut

Your servers. Your terminal. Your AI agent. One TUI.

Manage AWS, Hetzner, OVH, and custom servers from one terminal — with a built-in AI assistant and MCP server.

Quick Install

Linux / macOS:

curl -sSL https://raw.githubusercontent.com/zb-ss/servonaut/master/install.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/zb-ss/servonaut/master/install.ps1 | iex

Or install directly via pipx / pip:

pipx install servonaut

Manual install from source:

git clone https://github.com/zb-ss/servonaut.git
cd servonaut
pipx install .

Screenshots

Instance list — AWS, Hetzner, OVH, and custom servers merged into one view

Sidebar reveals Fleet Memory, Memory Sync, Secrets, Settings, and per-provider management for OVH and Hetzner

Built-in AI assistant with MCP server integration — chat with local providers or hosted Servonaut AI

CloudWatch log browsing with Top IPs analysis, geolocation, and abuse scoring

Ban/unban IPs via WAF, Security Groups, or NACLs with audit trail

All screenshots and the launch video were recorded with --demo active, which replaces real IPs, ARNs, paths, and secrets with safe fake equivalents. See docs/demo-mode.md for what is redacted and how to use it.

Features

• Interactive TUI with mouse and keyboard support powered by Textual
• Multi-provider — AWS EC2, OVHcloud (dedicated servers, VPS, Public Cloud), Hetzner Cloud (full lifecycle — list / create / destroy), plus custom servers from any provider (DigitalOcean, on-prem, etc.) with full SSH/SCP support
• List and search instances across all AWS regions with OVH and Hetzner instances merged into the same view
• SSH into instances — launches in new terminal window with auto-detected emulator
• Run remote commands via overlay panel with real-time streaming output, persistent history, and saved command favorites
• Per-instance dashboard — clicking an instance opens a Server Actions view with a sectioned action rail plus a detail pane showing an at-a-glance server-memory snapshot (OS, disk, web stack, databases, runtimes, containers) and an opt-in live resource monitor (CPU / RAM / load / disk / uptime, press L to start — polled over SSH only while open, never in the background)
• Browse remote file systems — interactive file tree navigation, available inline in the dashboard or as a full screen
• SCP file transfer — upload/download files and directories
• Real-time log viewer — stream remote logs via tail -f with pause, search, and log switching
• Keyword-based server scanning — search file contents across instances
• CloudTrail event browser — browse AWS CloudTrail events with filters for region, time range, event name, and user
• CloudWatch Logs browser — browse AWS CloudWatch log groups with Top IPs analysis, IP geolocation lookup, and AbuseIPDB integration
• IP ban manager — ban IPs via AWS WAF, Security Groups, or NACLs with audit trail
• OVHcloud management — OVH → ⚙ Manage per-provider screen with create / start / stop / reboot / delete (Cloud / VPS / dedicated routed automatically), region-first create wizard with API-backed flavor pricing, plus DNS zones, IP blocks and failover IPs, snapshots, block storage, billing and invoices, project-level SSH keys
• Hetzner Cloud management — Hetzner → ⚙ Manage per-provider screen with full lifecycle (create / power on / shutdown / power off / reboot / delete), state-aware action toolbar, project SSH-key registry, plus equivalent CLI (servonaut hetzner list / create / destroy / ssh-keys / server-types). Auto-registers new servers into the fleet. Full docs
• AI log analysis — analyze logs with OpenAI, Anthropic, Gemini, or Ollama (local install or Ollama Cloud) with cost estimation
• Built-in AI chat — LLM assistant with tool-calling against your instances (powered by the same MCP tool surface below)
• Servonaut AI — hosted AI gateway included with Solo and Teams plans. Subscribe at servonaut.dev and chat with your fleet without configuring any local API key. The model can tail logs, run commands (with confirmation), and triage incidents through the existing Mercure relay — your AWS credentials and SSH keys never leave the CLI. Quota and top-up balance are shown inline in the chat panel and via servonaut ai quota.
• Bring your own key — prefer to use your own model? Configure each cloud provider's key independently in Settings → AI Provider (ai_provider.openai_api_key, ai_provider.anthropic_api_key, ai_provider.gemini_api_key, ai_provider.ollama_api_key for Ollama Cloud). Local Ollama needs no key — just point ai_provider.base_url at your install. All options coexist with Servonaut AI; a one-time picker lets you choose the default and you can switch per-session from the chat-panel header.
• Server memory — persistent per-server cache of OS/runtime/service/web-stack/log/database/container/network/git/disk facts. Agents call get_server_memory(id) before SSH round-trips; CLI has servonaut memory build|refresh|show|export|annotate|pin|clear. Full docs
• Memory Sync — Solo+ feature that backs up your fleet memory to servonaut.dev with end-to-end encryption (X25519 keypair + AES-256-GCM envelopes wrapped to a passphrase you control). Drift detection across re-probes, cross-device history, and AI-queryable fact cache. The TUI's ☁ Memory Sync sidebar entry is the unified setup / unlock / status hub.
• MCP server for AI agents — Claude Code, Cursor, Windsurf, etc. ~60 tools covering instance ops, AWS EC2 lifecycle + describe helpers, S3 / object storage on AWS, Hetzner, OVH, AWS log analysis & IP banning (CloudWatch / CloudTrail / WAF / Security Group / NACL), full Hetzner + OVH lifecycle (create / start / stop / reboot / delete), SSH-key registry CRUD, server-memory queries, session introspection, and authenticated REST proxy. Three-tier guard system (readonly / standard / dangerous), confirmation-protocol prompt baked into every mutating tool's description, and a JSONL audit trail.
• Servonaut Cloud account — optional; run servonaut login (or TUI → Account → Login) to unlock config sync across machines and the MCP relay
• MCP relay — servonaut connect (or the TUI autostart) keeps a Mercure SSE connection open so AI agents and team-mates can dispatch MCP tool calls to this machine over the internet. Tokens never leave the CLI; heartbeats every 30 s with automatic Mercure JWT refresh.
• Config sync — client-side-encrypted snapshots of your config.json pushed/pulled from servonaut.dev, paired with a passphrase you control
• Bastion host / jump server support via ProxyJump or ProxyCommand
• Per-host SSH tuning — extra_ssh_options per connection profile / custom server for legacy boxes (HostKeyAlgorithms=+ssh-rsa, custom keepalives, etc.)
• SSH key management with auto-discovery and per-instance configuration
• Instance caching with stale-while-revalidate for fast startup
• Auto-update check — notifies of new versions on startup, one-click update from the menu or servonaut --update
• Desktop shortcut — servonaut --install-desktop adds an app launcher entry (Linux/macOS)
• Fully configurable — all settings in ~/.servonaut/config.json

Prerequisites

• Python 3.10+
• AWS CLI configured (~/.aws/credentials and ~/.aws/config)
• SSH client (standard on Linux/macOS, OpenSSH on Windows)
• pipx for isolated installation (recommended)

Your AWS credentials need ec2:DescribeInstances and ec2:DescribeRegions permissions. Additional permissions needed for optional features:

Getting Started

servonaut

That's the whole interface. The TUI is the primary and recommended way to use Servonaut — every feature (fleet view, SSH, remote commands, logs, IP banning, AI chat, server memory, provider management) is reachable from the sidebar, with full mouse and keyboard support.

A few flags you may want on day one:

servonaut --update                # Check for updates and upgrade
servonaut --install-desktop       # Create desktop shortcut (Linux/macOS)
servonaut --setup-ovh             # Guided OVHcloud credential setup
servonaut --debug                 # Verbose logging to stderr

Headless & automation: every major feature also has a scriptable CLI (servonaut connect, servonaut memory, servonaut ai, servonaut hetzner, servonaut secrets) for CI runners, cron jobs, and boxes without an interactive session — see the CLI Reference. Wiring up an AI agent instead? Jump to MCP Server for AI Agents.

Keyboard Shortcuts

What You Can Do

The TUI opens to a unified instance list (AWS + OVH + Hetzner + custom servers in one searchable table). The collapsible left sidebar groups everything else by purpose:

Core

• 📋 Instances — search and SSH the unified fleet
• 💻 Custom Servers — add / edit / remove non-AWS servers (DigitalOcean, on-prem, etc.)
• 🔑 SSH Keys — configure default and per-instance keys

Logs & Security

• 📊 CloudWatch — browse AWS log groups with Top IPs analysis, action filter (All/Allowed/Blocked), IP geolocation, AbuseIPDB lookup
• 🔒 IP Ban Manager — ban IPs via WAF, Security Groups, or NACLs
• 🔍 CloudTrail — audit AWS API activity with filters

Tools

• 🧠 Fleet Memory — scan / refresh / inspect the AI-queryable fact cache
• ☁ Memory Sync — encrypted backup of fleet memory across devices (Solo+)
• 🔄 Sync Config — encrypted config snapshots (Solo+)
• 🔧 Settings — configuration, scan rules, AI provider, AbuseIPDB key

OVH (visible when configured)

• ⚙ Manage — table of OVH instances with state-aware lifecycle toolbar (Create / Start / Stop / Reboot / Delete)
• 🔑 SSH Keys — project-level SSH key registry (the one the create wizard injects from)
• DNS Zones · IP Management · Block Storage · Billing

Hetzner (visible when configured)

• ⚙ Manage — table of Hetzner servers with full lifecycle toolbar (Create / Power on / Shutdown / Reboot / Delete)
• 🔑 SSH Keys — Hetzner Cloud project SSH key registry

Account

• Login · Teams · Bug Reports

Server Actions (clicking any instance row): a per-instance dashboard — the detail pane shows the server's identity, a memory snapshot, and an opt-in live resource monitor (L), while the action rail covers Browse Files (opens inline in the dashboard), Run Command, SSH Connect, SCP Transfer, View Scan Results, View Logs (tail -f), AI Analysis, Ban IP, and Manage/Verify SSH Ref

Command history persists across sessions — use Ctrl+R to search history and saved commands, Ctrl+S to save favorites.

Instance Caching

Configuration

All configuration lives in ~/.servonaut/config.json, created automatically on first run.

See Configuration Guide for the full reference including connection profiles, custom servers, scan rules, and match conditions.

Legacy / special-case SSH hosts: connection profiles and custom servers both accept an extra_ssh_options array that appends arbitrary -o KEY=VALUE flags per host — use it to talk to ancient OpenSSH boxes (HostKeyAlgorithms=+ssh-rsa), tune keepalives, or set connect timeouts without weakening your global SSH defaults. See Per-host SSH tuning.

Secrets: API keys in config.json support $ENV_VAR and file:~/.secrets/key syntax so the config file stays secret-free. You can also create ~/.secrets/servonaut.env with KEY=value pairs — loaded automatically on startup.

Optional Dependencies

# MCP server for AI agents
pipx inject servonaut mcp
# or: pip install 'servonaut[mcp]'

# Hetzner Cloud / OVHcloud provider SDKs
pip install 'servonaut[hetzner]'
pip install 'servonaut[ovh]'

# Install everything
pip install 'servonaut[all]'

AI log analysis (OpenAI, Anthropic, Gemini, Ollama) needs no extra install — httpx ships as a base dependency.

MCP Server for AI Agents

This section is for wiring up AI agents (Claude Code, Cursor, Windsurf, …) — not day-to-day interactive use. If you're a human operating your fleet, the TUI above is the recommended interface.

Servonaut includes an integrated MCP server that exposes tools to AI agents like Claude Code:

# Auto-install into a coding agent
servonaut --mcp-install claude     # Claude Code
servonaut --mcp-install cursor     # Cursor
servonaut --mcp-install windsurf   # Windsurf
servonaut --mcp-install opencode   # OpenCode
servonaut --mcp-install vscode     # VS Code Copilot
servonaut --mcp-install all        # All of the above

# Run MCP server manually (stdio transport)
servonaut --mcp

Agent-only / headless install

You don't need the TUI to use Servonaut as an agent toolbox. The MCP server runs fully headless — servonaut --mcp never loads the terminal UI (this is enforced by a regression test), so you can install it on a server or CI box purely as an MCP backend for your coding agent:

pipx install 'servonaut[mcp]'
servonaut --mcp-install claude   # or cursor, windsurf, opencode, vscode, all

Configure credentials and servers the same way as a TUI install ( ~/.servonaut/config.json, $ENV_VAR / file: secret syntax — see Configuration Guide). For Servonaut Cloud features (relay, config sync, hosted AI), servonaut login runs the device-flow sign-in fully headless — approve from a browser on any device. Everything an agent does goes through the same guard levels and is logged to ~/.servonaut/mcp_audit.jsonl.

Available tools:

The tool list is filtered to what's actually usable: OVH and Hetzner tools appear only when those providers are configured, the ip_ban_* tools only when at least one IP-ban target is defined, and the *_server_memory* tools only when the memory subsystem is enabled. CloudWatch/CloudTrail and the core instance tools are always available (AWS is the base provider).

• cloudwatch_top_ips parses WAF/ALB structured logs to rank client IPs with allowed/blocked counts — pair it with cloudtrail_lookup_events to corroborate, then ip_ban_set to block via WAF, a security group, or a NACL.
• whoami returns session metadata — the OAuth bearer is never exposed.
• api_request lets an agent make authenticated REST calls against servonaut.dev with automatic 401 refresh and a CLI-side rate limit (30/min). The bearer stays on the CLI.
• mcp_tool_call wraps a JSON-RPC 2.0 tools/call envelope against the hosted MCP at mcp.servonaut.dev — used for premium tools when your plan includes them.
• get_server_memory(id) returns the cached fact snapshot — agents call this BEFORE any SSH round-trip so they answer most OS / runtime / service questions without run_command. Pass format='context_block' to get back a <CONTEXT> envelope for direct prompt injection.

Guard levels: readonly (list/status/introspection only — includes CloudWatch/CloudTrail and ip_ban_list_* queries), standard (read + safe commands + authenticated REST + power management — start / stop / reboot / shutdown + S3 download), dangerous (everything, including create_server / delete_server / transfer_file / ip_ban_set / aws_terminate_instance / aws_run_instances / S3 mutations (s3_create_bucket, s3_delete_bucket, s3_upload_object, s3_delete_object, s3_copy_object, s3_move_object, s3_generate_presigned_url)). Dangerous shell commands (rm -rf, shutdown, reboot, etc.) are always blocked regardless of guard level. Mutating tools carry an explicit "confirm with the user before calling" cue in their descriptions; the top-level MCP instructions document the three-step protocol (summarise → state args → wait for affirmative reply). All operations are logged to ~/.servonaut/mcp_audit.jsonl.

Set Up with an AI Agent

Paste this prompt into Claude Code, Cursor, or any AI coding assistant to get Servonaut installed and configured automatically:

Install and configure Servonaut, a TUI for managing servers.

1. Install: `pipx install servonaut` (or `pip install servonaut`)
2. Install optional deps: `pipx inject servonaut mcp` (for the MCP server)
3. Run `servonaut` once to generate ~/.servonaut/config.json
4. Read ~/.servonaut/config.json and help me configure:
   - AWS regions to scan (default scans all, set `regions` array to limit)
   - Default SSH username (`default_username`, default "ec2-user")
   - Cache TTL (`cache_ttl_seconds`, default 3600)
   - Terminal emulator if not auto-detected (`terminal_emulator`)
5. If I use bastion/jump hosts, help me set up `connection_profiles` and `connection_rules`
6. If I have non-AWS servers, help me add them to `custom_servers`
7. If I want AI log analysis, help me configure `ai_provider` (openai/anthropic/gemini/ollama)
   - Each cloud provider has its own dedicated key field (`openai_api_key`, `anthropic_api_key`, `gemini_api_key`, `ollama_api_key`)
   - All key fields support `$ENV_VAR` and `file:~/.secrets/key` syntax so they don't go in the config file
8. Install MCP server into your coding agent: `servonaut --mcp-install claude` (or `cursor`, `windsurf`, `opencode`, `vscode`, `all`)

After setup, launch with `servonaut` and walk me through the key features.

Servonaut Cloud account

Optional — Servonaut works fully offline against your own AWS / OVH credentials. Signing in at servonaut.dev unlocks:

• Config sync — push/pull an encrypted snapshot of your config.json between machines. The passphrase never leaves your client; the server only sees ciphertext. Sidebar entry 🔄 Sync Config opens the snapshot manager directly (Pull Latest / Push New / Restore / Rename / Delete).
• MCP relay — a Mercure SSE channel that lets AI agents and team-mates dispatch MCP tool calls to this machine. While the relay is connected, https://servonaut.dev/account reports your CLI as online, and hosted agents can reach it.
• Memory Sync (Solo+) — encrypted fleet memory backup with drift detection. Open the ☁ Memory Sync sidebar entry, click Unlock Memory Sync, and enter a passphrase. The same screen handles first-time enrolment AND post-restart unlock — your private key is wrapped with the passphrase locally, so the server never sees it. After unlock, click Sync now to push every cached server's memory modules as encrypted envelopes. Per-feature settings (digest cadence, Mercure push, AI consent) live at the bottom of the Settings panel and are stored on your servonaut.dev account.

Sign in from the TUI's Account / Login screen. After a successful device-flow authentication, the TUI auto-starts an in-process relay listener and the sidebar indicator flips to ● connected.

The listener is tied to the TUI window — closing the TUI drops the connection after ~60 s. For always-on reachability (CI runners, headless boxes), use servonaut connect --bg instead; the CLI and TUI cooperate over ~/.servonaut/relay.lock so they can't run at the same time. The TUI shows external listener (PID N) when a --bg listener is holding the connection.

servonaut connect                 # Foreground relay (Ctrl+C to stop)
servonaut connect --bg            # Detach; writes ~/.servonaut/relay.pid
servonaut connect --status        # Local + backend view with divergence warning
servonaut connect --stop          # SIGTERM the background listener
servonaut connect --reconnect     # Heal a stale SSE socket (stop+start)
servonaut connect --force-bg      # Take over from a TUI's in-process listener

See CLI Reference → servonaut connect for full flags.

Tokens are stored at ~/.servonaut/auth.json with mode 0600, written atomically via tmp + os.replace(). If an older build left the file world-readable, the next run auto-fixes it.

Secrets management (Solo+)

Centralise SSH keys + other named secrets behind a pluggable provider backend. Day-to-day this is invisible: once configured, SSH key resolution checks your provider automatically on every connect — you keep clicking SSH in the TUI and it just works. The commands below are one-time setup. MVP supports two backends:

• LocalProvider — keys live in ~/.servonaut/secrets.json (mode 0600, atomic write, same trust model as auth.json). Always available on Solo and Teams plans.
• BitwardenProvider (bws) — keys live in your team's Bitwarden Secrets Manager project. Team admin configures the project from https://servonaut.dev/account/teams/<slug>/secrets; CLI fetches the metadata and reads/writes through the local bws binary using your own access token. The token never leaves your machine — servonaut.dev only stores the project ID and the name of the env var holding the token.

To use Bitwarden as your team's backend:

# 1. Install the bws CLI (one-time)
servonaut secrets install bws         # macOS: brew · Linux: cargo
# Windows / other → prints upstream install URL.

# 2. Mint a BWS access token (https://bitwarden.com/help/personal-access-tokens/)
#    and export it
export BWS_ACCESS_TOKEN=<your-token>

# 3. Verify wiring
servonaut secrets status              # shows plan, entitlement, active provider

# That's it — SSH key resolution now checks Bitwarden first, ~/.ssh as
# fallback. Push a key into BWS with `bws secret create`:
bws secret create "$(basename ~/.ssh/prod-server)" \
                  "$(cat ~/.ssh/prod-server)" \
                  --project-id <project-uuid-from-status>

Key resolution order on every SSH connect:

1. Active provider (Bitwarden, if configured) looked up by key name.
2. ~/.ssh discovery (existing patterns + fuzzy match).
3. The path stored in config.json::instance_keys[<id>] or config.default_key.

Free-tier users get the legacy ~/.ssh-only flow with zero behaviour change. Provider-supplied keys land in ~/.servonaut/keys/<name> at mode 0600.

Threat-model + design notes are pinned in the codebase via inline docstrings on services/secret_provider.py, services/bitwarden_provider.py, and services/secret_provider_resolver.py.

Development

# Run directly (primary dev workflow)
PYTHONPATH=src python3 -m servonaut.main

# Run with debug logging
PYTHONPATH=src python3 -m servonaut.main --debug

# Install editable
pip install -e .

# Update pipx installation after changes
pipx install . --force

# Run tests
pip install -e ".[test]"
pytest

See Architecture for codebase structure and design patterns.

Troubleshooting

See Troubleshooting Guide for help with SSH connections, bastion hosts, key management, and AWS credentials.

Runtime Files

All runtime files are under ~/.servonaut/:

Logging

Logs are always written to ~/.servonaut/logs/servonaut.log. Use --debug for verbose stderr output.

When SSH fails, the terminal window stays open showing the error and exit code.

License

This project is licensed under the MIT License — see the LICENSE file for details.