Skip to main content
A request is one call your backend makes through BundleUp using a connection: a Proxy call, a Unified API call, or an MCP message. Requests are the unit BundleUp logs, rate-limits and bills.

What gets logged

Every request is logged after the response is sent, so logging adds no latency. Each log entry records: For requests that reach the provider, the URL logged is the provider’s URL that BundleUp called. For requests BundleUp rejected before that, it’s the BundleUp URL you called. Request and response bodies are never logged. Neither are headers. What your users’ data contains stays out of BundleUp’s records. You can browse and filter logs in the dashboard under Logs. See Logs & observability.

Log retention

Older entries are deleted automatically. If you need a longer history, keep your own records of the calls you make.

What counts toward usage

Your plan includes a number of requests per month (see Pricing). A request counts each time BundleUp logs one, which includes:
  • Every Proxy, Unified API and MCP request, successful or not.
  • Every retry attempt. If BundleUp retries a GET twice before it succeeds, that’s three requests. Each attempt is a real call to the provider.
  • Every MCP message. initialize, tools/list, tools/call and notifications each count as one.
  • Requests BundleUp rejects after identifying your workspace, such as an invalid connection ID (400) or a rate-limited request (429).
These don’t count:
  • Requests with a missing or invalid API key (401). BundleUp can’t tell whose they are.
  • The authorization flow (users connecting accounts) and the code exchange.
  • Platform API calls (api.bundleup.io) and anything you do in the dashboard.
  • Webhooks BundleUp sends to you.

Rate limits

Each connection can make 100 requests per 60 seconds, counted across the Proxy, Unified API and MCP together. The limit is per connection, so one busy account never slows down another. See Rate limits.

Keeping usage down

  • Cache what doesn’t change often, like a user’s profile or a list of channels.
  • Use webhooks from the provider, where available, rather than polling.
  • Ask for bigger pages. One request for 100 items is cheaper than ten for 10. Unified API endpoints accept limit up to 100.
  • Avoid re-listing MCP tools every turn if your MCP client can cache the list.