MCP server

Safetybits runs an MCP server that gives an MCP client read access to your account: inventory, risks, vulnerabilities, alerts, detection rules, and network traffic. Point a client at it and ask questions about your OT network, backed by the same data the console shows you.

Connect a client

The server speaks the MCP authorization spec over HTTP, so a client that supports it, such as Claude Code or Claude Desktop, discovers how to authenticate on its own. Point it at the server URL and it fetches the server’s OAuth metadata, sends you to log in through your identity provider with your existing Safetybits account, and stores the resulting token itself. There is nothing to generate or copy in the console.

The OAuth client ID is safetybits-mcp, and the server requests the openid and email scopes. You don’t need to enter these yourself if your client handles discovery, but they matter if you’re configuring a client that asks for them explicitly.

The read-only guarantee

Every tool the server exposes is read-only and idempotent. None of them can create, modify, or delete anything in your account: they only retrieve data that already exists. That holds for every tool by construction, not by convention, so connecting an MCP client to a production account can’t change anything in it.

How scoping works

Safetybits accounts (plants) are scoped per user. If you don’t already know which account you’re working with, start with the accounts_find tool to list the accounts you have access to, then pass that account’s ID to every other tool. An organization administrator can reach any account in their organization; anyone else can only reach accounts they’ve been explicitly granted access to.