Bravado API Rate Limits, Backoff, and Retry Strategy

Sign each request with a fresh HMAC signature. The cURL and HTTP header values below are placeholders, not reusable credentials. For runnable signing code, use request signing examples. Python examples use the signed session from that page; Node.js examples use its bravadoFetch helper.

Bravado enforces rate limits on a per-API-key basis to ensure fair and stable access for all integrations. Your specific limit is tied to your account tier and is visible in your account snapshot, you can retrieve it programmatically at any time without hitting a separate billing dashboard.

Finding your rate limit

Call GET /v2/trade/account to retrieve your current API key metadata, including your rate limit:

GET /v2/trade/account HTTP/1.1
X-BRAVADO-API-KEY: <public-key>
X-BRAVADO-TIMESTAMP: <unix-ms>
X-BRAVADO-SIGNATURE: <signature-for-this-request>

The response includes an api_key object with a rate_limit_per_min field:

{
  "api_key": {
    "rate_limit_per_min": 120,
    "key_prefix": "bvd_live_****",
    "scopes": ["trade.read", "trade.write"]
  }
}

In this example, your key is permitted 120 requests per minute across all Bravado endpoints combined.

When you hit the limit

When you exceed your rate limit, Bravado responds with:

HTTP status: 429 Too Many Requests
Response header: Retry-After: <seconds>, the number of seconds you must wait before retrying
Response body: {"error": "rate limit exceeded"}

HTTP/1.1 429 Too Many Requests
Retry-After: 12
Content-Type: application/json

{"error": "rate limit exceeded"}

Never retry a 429 response immediately in a tight loop. Doing so will not succeed and may extend the window during which your key is rate limited.

Retry strategy

The recommended approach is exponential backoff with jitter, with the Retry-After header taking precedence over any computed delay. Python

import time
import random

def call_with_retry(fn, max_attempts=5):
    for attempt in range(max_attempts):
        resp = fn()
        if resp.status_code != 429:
            return resp
        # Honour Retry-After if present; otherwise back off exponentially
        retry_after = int(resp.headers.get("Retry-After", 2 ** attempt))
        jitter = random.uniform(0, 1)
        time.sleep(retry_after + jitter)
    raise Exception("Max retries exceeded")

JavaScript / Node.js

async function callWithRetry(fn, maxAttempts = 5) {
  for (let attempt = 0; attempt < maxAttempts; attempt++) {
    const resp = await fn();
    if (resp.status !== 429) return resp;

    const retryAfter = parseInt(resp.headers.get("Retry-After") ?? String(2 ** attempt), 10);
    const jitter = Math.random(); // 0–1 seconds
    await new Promise((resolve) => setTimeout(resolve, (retryAfter + jitter) * 1000));
  }
  throw new Error("Max retries exceeded");
}

Apply this wrapper to all mutating requests (POST, DELETE, PATCH). For read-only GET requests you can use a lighter strategy, a single retry after the Retry-After delay is usually sufficient.

Idempotency keys and safe retries

All mutating requests should include an Idempotency-Key header. If your request times out or you receive a transient error before you know whether it succeeded, you can safely resend the identical request with the same key, Bravado will deduplicate it server-side and return the original result without creating a duplicate order or withdrawal.

POST /v2/trade/order HTTP/1.1
X-BRAVADO-API-KEY: <public-key>
X-BRAVADO-TIMESTAMP: <unix-ms>
X-BRAVADO-SIGNATURE: <signature-for-this-request>
Idempotency-Key: 550e8400-e29b-41d4-a716-446655440000
Content-Type: application/json

Use UUID v4 values to guarantee uniqueness across requests.
The same key is safe to reuse only for the same logical operation. Generate a new UUID for every distinct request.
Idempotency keys are scoped to your API key, so there is no collision risk across accounts.

Idempotency keys protect against double-submission during network retries. They do not bypass rate limits, a deduplicated replay still counts as a request if the original already succeeded.

MCP server limits

The MCP server is limited per connection (your partner account plus the specific OAuth client), not per IP. AI clients reach Bravado through shared egress addresses, so a per-IP limit would pool unrelated customers into one bucket and let the busiest of them throttle everyone else. Two limits apply:

Limit	Applies to	On exceeding
Request rate	every MCP call	`429` with `Retry-After`
Query budget (bytes scanned per hour)	`run_sql` only	the call is refused with a message naming the budget

The query budget exists because request counts do not describe cost. A single query can read tens of gigabytes to return ten rows, and only a budget denominated in bytes scanned can bound that. The purpose-built tools are unaffected by it, so an agent that hits the budget can keep working through them. Every run_sql result reports how many bytes it actually read, so an agent can pace itself rather than discovering the limit by hitting it.

Best practices

Batch cancellations, use DELETE /v2/trade/orders (cancel-all) instead of looping individual cancel requests. A single call consumes one rate-limit slot regardless of how many open orders you have.
Use strategies for high-frequency fills, if you need many small orders placed at regular intervals, use the interval_sec field on execution strategies rather than calling the orders endpoint in a loop.
Monitor proactively, poll GET /v2/trade/account periodically to confirm your rate_limit_per_min hasn’t changed and to detect unexpected spikes in your usage patterns.
Share limits across threads, if your integration is multi-threaded or runs multiple processes, coordinate request counting centrally (e.g. via a token-bucket in Redis) to avoid one thread consuming the budget of another.

​Finding your rate limit

​When you hit the limit

​Retry strategy

​Idempotency keys and safe retries

​MCP server limits

​Best practices

Finding your rate limit

When you hit the limit

Retry strategy

Idempotency keys and safe retries

MCP server limits

Best practices