Should I rate-limit by IP or by user ID?

By authenticated user or API key. IP-based limiting can be bypassed with proxies and can unfairly throttle multiple legitimate users behind the same NAT.

What should a rate-limited response look like?

HTTP 429 with a Retry-After header, plus X-RateLimit-Limit and X-RateLimit-Remaining headers so well-behaved clients can back off correctly.

How do I rate limit across multiple server instances?

Move the counters into a shared store such as Redis so every instance checks and updates the same count for a given key.

complianceHowTo

MCP Rate Limiting Guide

Prevent abuse and ensure fair usage of MCP servers with token-bucket rate limiting, per-tier limits, and distributed limiting across multiple instances.

Quick Answer / TL;DR

Rate limit MCP servers with a token-bucket algorithm keyed by user or API key rather than IP address, since IP-based limits can be bypassed by proxies or unfairly affect multiple users behind the same NAT. Return HTTP 429 with a retry-after header, and use a shared store such as Redis once you run more than one server instance.

Key Takeaways

Token bucket allows short bursts while holding to an average rate, which fits real client behavior better than a fixed window.
Key limits by authenticated user or API key, not raw IP, for fairness and to resist bypass through proxies.
Once you run multiple instances, rate-limit state must live in a shared store such as Redis, not in local memory.

Token bucket limiter

A token bucket refills at a constant rate up to a cap and spends one token per request. This allows short bursts while still enforcing a long-run average rate, which suits typical MCP client traffic better than a hard fixed-window counter.

typescript

class TokenBucketLimiter {
  private buckets = new Map<string, { tokens: number; lastRefill: number }>();
  constructor(private maxTokens: number, private refillPerMs: number) {}

  isAllowed(key: string): boolean {
    const now = Date.now();
    const bucket = this.buckets.get(key) ?? { tokens: this.maxTokens, lastRefill: now };
    bucket.tokens = Math.min(this.maxTokens, bucket.tokens + (now - bucket.lastRefill) * this.refillPerMs);
    bucket.lastRefill = now;
    this.buckets.set(key, bucket);
    if (bucket.tokens < 1) return false;
    bucket.tokens -= 1;
    return true;
  }
}

Per-tier limits and distributed limiting

Different plan tiers should get different bucket sizes. Once the MCP server runs as more than one instance behind a load balancer, move the counters to Redis so every instance enforces the same limit against the same key.

Tier	Suggested limit
Free / readonly	10 requests/sec
Pro	100 requests/sec
Enterprise	1000 requests/sec