Usage policy & limits
Rate limits, backoff, and what the MCP server is not for.
30 requests per minute, per session
Every connected session gets 30 tool calls per minute, across all tools. The limit is keyed to the session, so one user's activity never affects another's.
Exceeding it returns:
{
"error": "RATE_LIMITED",
"message": "Too many requests, please try again later"
}
The response carries standard RateLimit-* headers and a Retry-After.
Back off on 429
A 429 means slow down, not fail. Honour Retry-After, then retry with exponential backoff and jitter. Retrying immediately turns one throttled call into a sustained block and makes the delay worse for the person waiting.
Where you can, show throttling to the user as a pause rather than an error.
Do not split a job across users to get more throughput
Spreading one bulk job across several users' connections to stay under each session's limit is a violation regardless of intent, and sustained patterns are treated as abuse.
If you need volume, use the REST API — it is built for batch generation, has limits sized for it, and does not require a person in the loop. MCP is for an assistant working alongside someone, a document at a time.
Act for the user, not around them
- Do not create, modify or export content the user did not ask for.
- Do not use a connection to copy a customer's template library, brand assets or datasets out of Marq.
- Be honest in your product about what was generated.
Updated 3 days ago