Guardrails

Apply policies to MCP requests.

Guardrails apply policies to MCP requests before and after they execute.

Available now

Rate limiting. Limit requests and token consumption per API key, server, or tool. See MCP Rate Limits.

Guardrail checks. Run the same guardrail checks available for LLM requests against MCP tool calls, before and after they execute. Denied calls return a structured error containing the guardrail and check ID.

Per-server guardrail mappings

Guardrails can be attached directly to individual MCP servers through the guardrail-to-server mapping API. Each mapping specifies:

• run_on — when the guardrail runs: beforeToolCall, afterToolCall, or both
• capability_ids (optional) — limit the guardrail to specific tools on the server. When omitted, the guardrail applies to all tools

Configure mappings via the Admin API or the SCM API using the guardrail MCP server mapping endpoints. Guardrails can also be configured at the workspace level using the standard guardrails workflow.

Creating MCP-targeted guardrails

When creating a guardrail intended for MCP servers, set target to mcp. The guardrail list API supports filtering by target (llm, mcp, or both). Guardrails with target mcp only appear when you explicitly filter for them.

What’s coming

Approval workflows. Require explicit approval for high-risk operations like deletions or write actions.

Budget controls. Limit usage based on cost or request volume.

Get early access

Contact us if you’re interested in early access to approval workflows or budget controls.

---

Portkey is now PRISMA AIRS AI Gateway. See it in action.

Contact Us