Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.breakcold.com/llms.txt

Use this file to discover all available pages before exploring further.

MCP tool calls use the same protected Breakcold API behind the scenes. This means MCP clients are rate limited just like REST API integrations. An AI assistant can make several tool calls from one user message. For example, one prompt might search records, open a record, read notes, and then update a task. Each tool call counts separately.

​
How limits work

Breakcold checks several limits at the same time. A tool call is blocked when any one of those limits is full.
LimitCurrent budget
Organization600 requests per minute
Workspace300 requests per minute
MCP application or agent identity180 requests per minute
Route120 requests per minute for the same route in the same organization
The limit that matters is the first one you reach. A busy AI tool can hit the MCP application or agent limit even when the rest of the organization is still below its limit.

​
What happens when MCP is rate limited

When an MCP request is rate limited, Breakcold returns HTTP 429. MCP clients receive a JSON-RPC error with code -32029 and message Public API rate limit exceeded. When possible, Breakcold also sends a Retry-After header in seconds. 
{
  "jsonrpc": "2.0",
  "id": "request_id",
  "error": {
    "code": -32029,
    "message": "Public API rate limit exceeded"
  }
}
Some AI tools show this as a normal error message. Others hide the technical details and only tell you to try again later.

​
What users should do

If your AI tool says it was rate limited, wait a moment and try again. To reduce repeated rate limits:
  • Ask for smaller tasks.
  • Avoid asking the assistant to scan everything repeatedly.
  • Let long-running workflows finish before starting another large one.
  • Use the same region URL as your workspace.

​
What MCP client builders should do

If you build or configure an MCP client, respect the Retry-After header when it is present. Good client behavior:
  • Wait before retrying a rate-limited tool call.
  • Do not retry many tool calls at the same time.
  • Cache recent reads when it is safe to do so.
  • Break large jobs into smaller batches.
  • Show a clear message such as “Breakcold is asking us to slow down. Try again in a moment.”