MCP Server
Give AI agents like Claude Code direct access to your servers — securely and with fine-grained control.
What is MCP?
Model Context Protocol (MCP) is an open standard created by Anthropic that lets AI assistants connect to external tools and data sources. Think of it as a plugin system for AI agents.
Servonaut ships with a built-in MCP server. When you connect an AI assistant to it, the assistant gains the ability to list your servers, fetch logs, run commands, and more — all through natural language conversation.
You can ask Claude Code: "Check the nginx error logs on prod-web-1 and tell me if there are any 5xx errors in the last hour." Claude calls Servonaut's MCP tools, fetches the logs, and gives you an analysis — without you writing a single command.
Quick install
The fastest way to connect Servonaut to Claude Code is the auto-installer. It modifies Claude Code's MCP configuration file automatically.
The MCP server requires mcp and httpx. Install both with pipx inject servonaut httpx mcp before running --mcp-install.
Manual setup for Claude Code
If you prefer to configure manually, add the following to your Claude Code MCP configuration
(usually at ~/.claude/claude_desktop_config.json):
After saving the file, restart Claude Code. The Servonaut tools will appear in the tools panel.
Available tools
The following MCP tools are exposed by Servonaut. The tools available depend on your configured guard level.
| Tool | Guard level | Description |
|---|---|---|
list_instances |
readonly | List all servers (AWS EC2 + custom). Returns name, IP, tags, and status. |
get_server_info |
readonly | Return detailed information about a specific server including instance type, region, and uptime. |
check_status |
readonly | Check SSH connectivity and basic health metrics (CPU, disk, memory via top / df). |
get_logs |
readonly | Fetch log file contents from a remote server. Supports line count, offset, and pattern filtering. |
run_command |
standard | Execute a shell command on a remote server via SSH and return stdout / stderr. Commands are logged. |
transfer_file |
standard | Upload or download a file to/from a remote server via SCP. |
manage_waf |
dangerous | Add or remove IP rules from AWS WAF, Security Groups, or NACLs. |
reboot_instance |
dangerous | Reboot an EC2 instance or send a remote reboot command to a custom server. |
Guard levels
Guard levels let you control how much power AI agents have over your infrastructure.
Set the level in config.json.
Read-only access. AI agents can list servers, fetch logs, and check status, but cannot modify anything. Safe for shared environments or untrusted agents.
list_instances, get_server_info, check_status, get_logs
Allows read operations plus safe writes: running commands and transferring files. Commands are logged. Recommended for personal use with trusted AI agents.
run_command, transfer_file
Full access including destructive operations: WAF management, instance reboots, and more. Use only in fully trusted, personal setups.
manage_waf, reboot_instance
Start with readonly and increase permissions only as needed.
The dangerous level should only be used in trusted, single-user setups.
For team environments, use the Teams plan
which adds MCP policy management and per-user permission control.
Using with other AI agents
Servonaut's MCP server uses the standard stdio transport and is compatible with any MCP-capable AI assistant, not just Claude Code. Tested integrations include:
- Claude Code (Anthropic) — first-class support via
--mcp-install - Claude Desktop — same config format as Claude Code
- Any MCP-compatible client — use
servonaut --mcp-serveras the server command
Premium: hosted MCP server
The Solo and Teams plans include a cloud-hosted MCP server. Instead of running the MCP server on your local machine, Servonaut's cloud connects to your servers on your behalf — no local installation required.
Hosted MCP benefits:
- Works from any device — AI agents connect to the cloud, not your laptop
- Always-on availability (not dependent on your machine being running)
- Teams plan adds MCP policy management and per-member access controls
- Audit trail of all MCP tool calls