Guard Rails

Work in progress, except API Credentials. API Credentials is a fully working feature that happens to live on this tab. Input Constraints, Output Filtering, and Tool Restrictions are visible and interactive, but nothing you enter in them is saved yet — treat those three as a preview, not working settings.

This tab holds two different things: the real, working API Credentials panel, and a preview of a safety layer (Input Constraints, Output Filtering, Tool Restrictions) that will eventually constrain what the AI can send and receive through a server's tools.

What you can do

• API Credentials (fully working) — configure the credentials Arthur attaches to every outgoing request when it calls your upstream API on behalf of a tool call: Bearer token, API key (header or query), Basic auth, OAuth2 client credentials, or custom headers. See How to Configure API Credentials for the full guide.
• Input Constraints (not yet functional) — meant to validate or reject tool call parameters before they reach the upstream API: block forbidden values, enforce allowed ranges, or require specific formats.
• Output Filtering (not yet functional) — meant to scrub sensitive data from tool responses before they reach the AI: mask fields, remove fields, or redact a regex pattern.
• Tool Restrictions (not yet functional) — meant to block a tool outright, cap how many times it can be called per session, or require confirmation before it runs.

How to use it

1. Open a server and go to its Guard Rails tab.
2. Configure API Credentials for real — this is what Arthur actually sends to your upstream API.
3. You can fill in Input Constraints, Output Filtering, or Tool Restrictions and click save, but the save will fail silently or show an error — none of it is persisted or enforced by the backend yet.

Good to know

• Don't confuse API Credentials with Access Keys or OAuth Client on the Connect tab — those protect incoming calls from an AI client to your MCP server. API Credentials is the outgoing side: what Arthur sends to your upstream API. See Secrets and Authentication.
• Related: Harness, the reliability-focused counterpart to Guard Rails's safety focus — most of it is in the same early state.