> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nimt.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Usage limits

> Request limits, pagination, and safe retry behavior.

Your AI assistant usually handles requests for you. These limits matter when
asking it to fetch many projects or building your own MCP client.

## Request rate

| Limit                          | Current allowance     | Shared by                                                           |
| ------------------------------ | --------------------- | ------------------------------------------------------------------- |
| Authenticated requests         | 120 per minute        | All OAuth connections and PATs for the same Nimt user and workspace |
| Requests before authentication | 600 per minute per IP | Clients using the same public IP address                            |

Both limits apply. Discovery requests such as `tools/list` also count, so one
chat message can use several requests. Creating another token for the same user
and workspace does not increase the authenticated allowance. OAuth sign-in and
token endpoints have separate limits.

If a limit is reached, Nimt returns HTTP `429` with a `Retry-After` header.
Wait at least that many seconds before retrying. Avoid rapid retries or repeatedly
reconnecting. For large inventories, use pagination and reduce parallel requests.

These are service protection limits, not a promise of plan eligibility or AI
client usage. ChatGPT and Claude apply their own subscription and usage limits.

## Page size and request size

* Project listing defaults to 25 projects and accepts up to 100 per call.
  Use the returned pagination details to fetch the rest.
* Request bodies are limited to 8 KiB. Send only the tool's documented inputs;
  do not attach conversation history or files to a tool call.
* The server expects uncompressed JSON over Streamable HTTP. It does not expose
  a local stdio server or a standalone SSE endpoint.

## Timeouts and retries

The MCP HTTP request deadline is 25 seconds. If a request times out or the
connection closes, reduce the size or frequency of requests and retry with a
bounded delay. Contact support if the problem persists.

The current two tools are read-only, so retrying them does not change project
data. A timeout does not generally mean a server operation was rolled back;
revisit retry behavior before using any future write tools.

For a custom client, retry temporary network failures and `429` responses with
bounded backoff. Reauthenticate for expired credentials. Do not repeatedly retry
invalid inputs or missing permissions without correcting the cause.

See [available tools](/mcp/tools#pagination) for pagination and
[troubleshooting](/mcp/troubleshooting#get-help) for what to include in a support request.