# Rate limiting

REST and MCP request limits, response headers, and retry behavior

Source: https://docs.rezzy.dev/rate-limiting
Documentation index: https://docs.rezzy.dev/llms.txt



Request limits are separate from generation credits. Status reads and documentation tools cost no generation credits but still count toward request limits.

## REST limits

REST requests pass through three layers in order: IP address, API key, then account. A limit at any layer returns HTTP 429.

| Layer      | Free        | Pro         | Window     |
| ---------- | ----------- | ----------- | ---------- |
| IP address | 60 requests | 60 requests | 60 seconds |
| API key    | 10 requests | 20 requests | 60 seconds |
| Account    | 60 requests | 60 requests | 60 seconds |

The account limit is shared by keys belonging to the same account. Requests with missing or invalid credentials still count toward the IP limit; a supplied bearer key also counts toward its key limit. The account layer applies when the key identifies a valid account.

## MCP limits

The [MCP endpoint](https://docs.rezzy.dev/mcp) uses separate windows from REST:

| Layer      | Limit        | Window     |
| ---------- | ------------ | ---------- |
| IP address | 600 requests | 60 seconds |
| API key    | 600 requests | 60 seconds |

MCP limits are the same across plans and do not have an account layer. Protocol requests, tool discovery, documentation reads, and status polling count toward these limits. `OPTIONS` requests are excluded.

## Response headers

| Header                  | Description                                       |
| ----------------------- | ------------------------------------------------- |
| `X-RateLimit-Remaining` | Requests left in the last layer checked           |
| `X-RateLimit-Reset`     | Unix timestamp in seconds when that window resets |
| `Retry-After`           | On HTTP 429, seconds to wait before retrying      |

These headers report the last layer checked, rather than every layer's remaining quota.

## HTTP 429 response

REST and MCP requests rejected by the rate limiter receive this HTTP JSON body before any tool executes:

```json
{
  "success": false,
  "message": "Too many requests. Please try again later.",
  "data": {
    "error_code": "RATE_LIMIT_EXCEEDED",
    "details": { "retry_after": 45 }
  }
}
```

The response includes `Retry-After`, `X-RateLimit-Remaining: 0`, and `X-RateLimit-Reset`. Honor `Retry-After`, poll generation status no faster than every five seconds, and back off between reads. Do not automatically repeat a creation request after an ambiguous connection failure: it can create and charge for a second document.
