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

# How MCP works

> Overview of AdCue MCP — tools, router pattern, and safety model.

# How MCP works

## Endpoint

```text theme={null}
https://mcp.adcue.app/mcp
```

Production serves the **V2** Google Ads tool surface (top-level helpers + `google_ads` router).

## What MCP gives you

AdCue exposes Google Ads operations as MCP tools so AI clients can:

* Read account and campaign state
* Create and manage Search campaigns (with safety rails)
* Optimize keywords, budgets, and wasted spend
* Reach advanced campaign types (Performance Max, YouTube, Demand Gen) when needed — marked **advanced** in the catalog

Tool **names and counts** come from the V2 registry export (`docs-site/generated/tools.json`).
See the [tool catalog](/mcp/tools) for the full list. Do not hand-edit tool lists in narrative pages.

## Router pattern

Most Google Ads operations go through the `google_ads` router.

List tools:

```json theme={null}
{ "action": "list_tools" }
```

Execute a tool:

```json theme={null}
{
  "action": "execute",
  "tool_name": "list_campaigns",
  "arguments": { "customer_id": "1234567890" }
}
```

Discovery helpers at the top level:

| Tool                      | Role                                        |
| ------------------------- | ------------------------------------------- |
| `search_tools`            | Natural-language discovery over the catalog |
| `get_tool_schema`         | Exact parameters for one tool               |
| `get_connections_status`  | Connection health                           |
| `list_connected_accounts` | Connected Google Ads customers              |
| `get_usage_status`        | Plan / quota summary                        |
| `switch_primary_account`  | Set the active customer for later calls     |

Some tools are also available **top-level** (not only via the router). The catalog marks each name’s exposure as `top`, `router`, or `both`.

## How a request flows

1. You type a goal in the AI client (or call a tool directly).
2. The client selects an AdCue tool and arguments.
3. AdCue authenticates your session / token and loads the stored Google Ads connection.
4. Read tools return data; write tools require product safety rules and your approval in the client.
5. The client presents the result.

## Safety (summary)

| Rule                | Detail                                                     |
| ------------------- | ---------------------------------------------------------- |
| Creates are paused  | New entities land **PAUSED**                               |
| String customer IDs | Always pass `customer_id` as a string                      |
| Confirm deletes     | Destructive `remove_*` requires `confirm_delete: true`     |
| Asset TTL           | Asset bundles expire in **1 hour**                         |
| Budgets             | Skill-facing budgets use **dollars** (decimal), not micros |

Full page: [Safety model](/mcp/safety).

## Advanced tools

A subset of tools for Performance Max, YouTube, and Demand Gen is flagged **advanced** in the generated catalog. Prefer core Search / account tools for everyday workflows unless you intentionally need those campaign types.

## Related

* [Connect an AI client](/connect/mcp-setup)
* [Tool catalog](/mcp/tools)
* [Safety model](/mcp/safety)