MCP servers

Give agents access to approved tools

RytherOS can connect an AI agent to a trusted Model Context Protocol (MCP) server. The server exposes a set of tools, and you choose which of those tools the account and each agent may use.

Connect a tool set, then narrow access

Custom Functions define actions one at a time. An MCP server can present multiple tools through one connection. RytherOS lets you test the connection, review discovered tools, disable tools account-wide, and further narrow access for a specific agent. Set execution limits before a tool becomes part of a live workflow.

Use trusted sources

An external server can receive information needed for a tool call. Review the server, its permissions, and the data it will handle before connecting it. Start with a narrow task and a clear human fallback for failures or unexpected results.

Frequently asked question

Can an agent turn on a tool that the account disabled?

No. Agent-level access can be narrowed but cannot exceed the account-level tool selection.

Review a trusted tool connection.

Bring the tools and handoffs your team already uses. We can map a practical next step around your workflow.

Talk with RytherOS