Rate Limit MCP Tools

Not all tool calls have equal cost. A metadata lookup and a 30-minute analysis job should not consume the same quota.

Useful dimensions

• Authenticated user/service.
• Tenant.
• Tool name.
• Concurrent calls.
• Downstream API quota.
• Compute or monetary cost.

Return actionable failures

When possible, tell clients whether the limit is temporary and when retry is reasonable. Avoid automatic retry storms.

Separate abuse limits from product quotas

Security rate limiting protects infrastructure; subscription quotas enforce commercial policy. They may share technology but should remain conceptually separate.