---
title: MCP Limits & Data Handling
description: How many requests an AI client may make over MCP, and what the client, its model provider and Apptonomy each receive and keep.
---

## Data handling

- **What the client receives** is the results of the [tools](/docs/mcp/tools) and [actions](/docs/mcp/actions): your project, audit, recommendation, score, revision and impact data, within the grant. Personal data (names, emails, user ids) is included only when the grant allows **Personal data**.
- **What happens to it next is up to the AI client you use.** The client and its model provider receive the results and handle them under your agreement with them — for example, whether conversations are retained or used for training. Check your AI provider's data settings, and your company's policy on which AI tools may receive company data.
- **What Apptonomy records** for each call: who called, from which client, the tool, the organization and project it named, whether it was allowed, and when. Apptonomy never records your sign-in token, your prompts, the client's replies or the data returned. Owners and Managers can see and export this under **Access history** on the API Access page; entries are kept for 90 days.
- Text that someone wrote — a project name, a note — reaches the client only as data inside a result, never as part of a tool's instructions.

More on how Apptonomy protects your data: [Security](/security), the [subprocessor list](/subprocessors) and the [DPA](/dpa).

## Limits

Each connection may make up to 120 requests a minute, and calls into an organization also count against that organization's limits, which it shares with its API keys (see [Rate limits and quotas](/docs/api/conventions#rate-limits-and-quotas)). A limited call returns a tool error with `rate_limited` or `quota_exceeded` and `retryAfterSeconds`; the connection stays open. A single result is capped at 150,000 characters: a page that would be larger is refused with `result_too_large` — ask for a smaller `limit`.
