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

# MCP Overview

> Connect an AI client to Alphractal and let it read the platform's metrics, charts, screener, derivatives and research as you.

Alphractal speaks the [Model Context Protocol](https://modelcontextprotocol.io). Point a
compatible AI client at our server and it can work with the same data the platform serves —
as you, on your plan, against your limits.

<Note>
  The MCP server is a different surface from the REST API documented here, but the same
  account: the key you create for the API authenticates the MCP server, and both draw on the
  same credit balance.
</Note>

## What it exposes

|               |                                                                                                                                                                                                           |
| ------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Tools**     | Assets and prices, on-chain and derivatives metrics, the chart catalog, the screener, liquidation clusters, trading indicators, research reports, and your own alerts, favorites and saved screener views |
| **Resources** | The metric catalog, the asset catalog, your favorites and the latest research — pullable by name                                                                                                          |
| **Prompt**    | `asset-deep-dive`, which runs the tools in an order that works                                                                                                                                            |

The full list is in [Tools and Resources](/mcp/tools).

## Why use it

**It is your account, not a public feed.** The client authenticates as you. What it can read
is what your plan entitles you to — the same tier rules as the web app and the REST API. No
separate quota, no separate billing.

**No integration code.** Instead of writing a client for 414 endpoints, you point the AI at
one server and ask questions in words. Resolving an asset, finding the right metric, picking
a timeframe and comparing series are tool calls the model makes on its own.

**Refusals are legible.** A tool above your plan says so in words, with the plan that would
lift it — it never returns empty data that reads as "nothing found". A rate-limited call says
how long to wait. An agent can act on both.

## What you can do with it

<CardGroup cols={2}>
  <Card title="Research an asset end to end" icon="magnifying-glass">
    Price and market summary, on-chain state, derivatives positioning and the latest research
    on one asset, gathered in a single conversation.
  </Card>

  <Card title="Screen and rank" icon="table-list">
    Run the screener, rank assets by any metric, and ask what moved today — then save the view
    for later.
  </Card>

  <Card title="Test an idea against history" icon="chart-line">
    Backtest a metric, compare two metrics or two charts, and pull the price history behind
    the result.
  </Card>

  <Card title="Manage your own alerts" icon="bell">
    List, edit and delete alerts, and create new ones through a two-step confirmation.
  </Card>
</CardGroup>

### Example prompts

```text theme={null}
What does BTC's on-chain state look like right now — supply in profit, MVRV and exchange flows?

Compare ETH and SOL open interest over the last 90 days at 4h resolution.

Which assets moved the most today, and is there research covering any of them?

Backtest MVRV z-score on BTC: what happened after it crossed above 3?

Create an alert for when BTC funding rate on binance goes negative.
```

## What it will not do

* **No tool reaches outside Alphractal.** Wallet and chain operations are deliberately not
  exposed.
* **Creating an alert takes two steps.** The agent stages the alert and gets back a token
  plus the exact configuration; only after you approve does a second call create it. What is
  created is what you were shown, whatever the model restates.
* **Anything that changes state** — alerts, favorites, saved screener views — is flagged as
  such in the protocol, so a well-behaved client asks you before running it.

## What it costs

The same two limits you already have:

* **API credits** — tools that fetch market data draw on the balance described in
  [API Credits](/credits). One tool call can fan out to several data requests, so a
  conversation spends faster than a script does.
* **Requests per minute** — your plan's [rate limit](/rate-limits). Over it, a tool answers
  `quota_exceeded` with how long to wait.

`get_my_plan` and `get_my_usage` report both. An agent that reads them before proposing work
will not walk you into a wall.

Ready to connect? See [Connecting to Alphractal MCP](/mcp/connecting).