# Overview

Pluggy is an open finance platform for Latin America. One API integration reaches
hundreds of financial institutions — bank accounts, credit cards, investments, loans,
identity and payments — and returns their data in a single, normalized shape, whether
it came from a regulated Open Finance connector or from one of Pluggy's own.

## The three objects

Almost everything in the API is one of these, and the rest hangs off them:

- **Connector** — one financial institution, and the products it can return.
- **Item** — one user's connection through a connector, created after they consent.
  It is the entry point to their data and the thing you store on your side.
- **Product** — the normalized data an item gives you: accounts, transactions,
  investments, identity, loans.

The [Glossary](/docs/get-started/glossary) has the rest of the vocabulary; the
[Item](/docs/connections/item) guide covers the lifecycle you will actually spend
your time on.

## Two ways to build the first integration

Both end in the same place — an item created and its transactions on your screen.

<Cards>
  <Card title="Build it with an agent" href="/docs/quickstart#build-it-with-an-agent" eyebrow="Fastest">
    Connect your coding agent to these docs with one URL, then hand it the prompt
    that builds the whole flow.
  </Card>
  <Card title="Follow the steps yourself" href="/docs/quickstart#or-do-it-by-hand">
    Five steps with curl: an application, an API key, a connect token, the widget,
    the data.
  </Card>
</Cards>

### Point an agent at the live docs

If you code with Claude, Cursor, Copilot, Codex or Gemini, connect them to our
documentation first. It is one URL, hosted, no key:

```bash
claude mcp add --transport http pluggy-docs https://mcp.pluggy.ai/mcp
```

The agent then reads the current guides and OpenAPI spec at runtime instead of
recalling whatever version of our API was in its training data — the difference
between code that runs and code that looks right. Setup for every other client is on
the [MCP server](/docs/developer-tools/mcp) page, and the prompt that builds the
whole flow is in the [Quick Start](/docs/quickstart).

Two more surfaces, for agents you cannot configure: every page here has a markdown
twin — add `.md` to its URL — and [`/llms.txt`](https://v2.docs.pluggy.ai/llms.txt)
indexes all of them. Anything that can fetch a URL can read our docs properly.

### Or follow the steps yourself

The [Quick Start](/docs/quickstart) is five steps: create an application, get an API
key, issue a connect token, open the widget, read the data. Nothing on this site
assumes you use an agent.

## Sandbox first

New applications start with access to the [Sandbox connectors](/docs/guides/sandbox),
so you can run the entire flow — including the failure cases you would rather find
now than in production — against fake institutions, with no real credentials
involved.