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

# Add a site to your agent

> Give your own product's agent one site's tools through that site's integration MCP.

Connect your product's agent to one site's integration MCP, instead of the Pomerado MCP, to give it only that site's tools. This guide uses a Google Flights integration.

## Get the URL and a key

On the Dashboard, open the Google Flights integration. Its Connect panel gives the MCP URL, and Create a key makes an integration key, which reaches only this integration's tools and jobs. Keep the key in your secret manager.

## Connect your agent

Connect your agent's MCP client over Streamable HTTP, with the key as a bearer token:

```json theme={null}
{
  "mcpServers": {
    "google-flights": {
      "type": "http",
      "url": "https://mcp.pomerado.ai/mcp/integrations/int_0f8e2d1c4b3a49e8a7f6e5d4c3b2a190",
      "headers": { "Authorization": "Bearer ${POMERADO_API_KEY}" }
    }
  }
}
```

## Call its tools

The MCP lists the site's tools by their own names, such as `search_flights`, plus `get_job`, `answer_job` and `cancel_job`. A site tool takes its input as an object:

```json theme={null}
{
  "name": "search_flights",
  "arguments": {
    "input": {
      "origin": "SFO",
      "destination": "JFK",
      "departure_date": "2026-11-20"
    }
  }
}
```

The call returns the job, usually finished in seconds. If it's still running, follow its `next` (`get_job` with `wait_seconds`).

## Handle a question

A job can stop with status `needs_input` and an `input_request`, which holds a message, its questions and an `answer_url`.

For a plain question, such as a choice, your agent asks the user, then calls `answer_job` with an answer for each question ID:

```json theme={null}
{
  "job_id": "job_6f1d0c2b3a4e45f8a9b7c6d5e4f3a2b1",
  "input_request_id": "req_0f8e2d1c4b3a49e8a7f6e5d4c3b2a190",
  "answers": "{\"cabin\": \"economy\"}"
}
```

For a login, a code or another secret, give the user the `answer_url`, a protected page that keeps the value out of your agent's conversation. Then keep waiting with `get_job`.

Either way the same job continues. Never start it again.

<Accordion title="Details">
  * Answer before the request's `expires_at`, or the job fails.
  * An integration key can run tools and follow, answer and cancel jobs. Building a tool or managing a login needs an account key.
  * More in [integration MCPs](/guides/integration-mcps/overview), [answer questions](/guides/jobs/answer-questions) and the [integration MCP reference](/mcp-reference/integration-mcp).
</Accordion>


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