> ## 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.

# Use Pomerado in chat

> Use the Pomerado MCP in a chat client to look up and compare data across a site, and to build new tools.

Use the Pomerado MCP in a chat client as a personal assistant for tasks on websites. This guide uses Instacart.

## Connect

In a chat client that supports custom MCP connectors, add `https://mcp.pomerado.ai/mcp` and sign in. To try it without an account, add `https://mcp.pomerado.ai/mcp/keyless` instead. See [client setup](/guides/connect/client-setup).

## Compare prices across stores

Ask a question that needs a lot of data at once:

> Compare the prices of eggs, oat milk, bananas and coffee at the Instacart stores near 94103, and tell me where my list costs least.

The agent finds an Instacart search tool with `find_website_tool`, then calls `run_read_only_website_tool` once for each item, all at the same time:

```json theme={null}
{
  "tool_id": "tool_0f8e2d1c4b3a49e8a7f6e5d4c3b2a190",
  "input": "{\"query\": \"oat milk\", \"zip\": \"94103\"}"
}
```

Read-only runs change nothing on the site, so they can run in parallel. Each usually returns in seconds, and the agent combines the results into one answer.

A search that needs no Instacart login also works on the keyless connection, within its daily limit.

## Build a missing tool

When no tool covers a task, ask the agent to build one:

> Build a tool that lists my past Instacart orders.

The agent calls `build_website_tool`, tells you it's building, and follows the job. A build takes about 10 minutes and needs you to be signed in to Pomerado. If it needs your Instacart login, the agent gives you a link to a protected page: enter the login there, never in chat.

## Track the build

* **In chat:** the agent calls `get_job` with `wait_seconds`. Each call returns when the job asks a question or finishes, or when the wait ends; the agent calls again while it runs.
* **On the Dashboard:** open the job's `watch_url`.
* **Elsewhere:** create a webhook on the Dashboard under Settings, then Webhooks, to be told when a job asks a question (`job.needs_input`) or finishes (`job.succeeded`, `job.failed`).

When the job succeeds, its `tool_id` names the new tool, and the agent can run it right away.

<Accordion title="Details">
  * How long one call can wait depends on the client. See [wait for a result](/guides/jobs/wait-for-a-result).
  * Clients that support MCP events get the same job events without a webhook. See [MCP events](/guides/notifications/mcp-events).
  * Set up a webhook in [webhooks](/guides/notifications/webhooks).
</Accordion>


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