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

# Risk tiers

> Understand read, write, and destructive MCP tool labels and how to configure client confirmations.

Every MCP tool is annotated with a **risk tier** so agents and humans can reason about side effects before a call runs.

| Tier | Description | Examples |
| - | - | - |
| `read` | Read-only; no data changes | `search_contacts`, `get_credits`, `list_campaigns` |
| `write` | Creates or modifies data | `research_contacts`, `create_campaign`, `send_email` |
| `destructive` | Permanent deletion | `delete_campaign`, `delete_list`, `delete_saved_search` |

## MCP client hints

Tools also expose MCP SDK hints derived from the tier:

* `readOnlyHint: true` for `read` tools
* `destructiveHint: true` for `destructive` tools
* `openWorldHint: true` when the tool touches unbounded external data (e.g. global search graph, sending email to the public internet)

## Recommended client settings

<Tip>
  Enable human confirmation for `write` and `destructive` tools in your MCP client. Read tools are safe to run autonomously for most workflows.
</Tip>

<Warning>
  `destructive` operations cannot be undone. Require explicit approval before `delete_campaign`, `delete_list`, or similar tools.
</Warning>

## Tool reference

Each [tool domain page](/mcp/tools/search) lists tools with their risk tier. See [Access control](/mcp/access-control) for which domains your account can use.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.