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

# Overview

> A fully hosted Model Context Protocol server that connects AI assistants, agent frameworks, and your own products to Blockworks onchain data and crypto intelligence.

Blockworks MCP is a fully hosted [Model Context Protocol](https://modelcontextprotocol.io) server. It gives AI
assistants and agents (Claude, Claude Code, Cursor, ChatGPT, or anything else that speaks MCP) direct access to
Blockworks market data, onchain metrics, and crypto research.

Unlike MCP servers that ship as a local package, there is nothing to install or run. You add one URL to your client
and authenticate in the browser with your Blockworks API key.

It is also the recommended way to build a product on Blockworks data, not just to equip an assistant you use yourself.
See [Building on MCP](#building-on-mcp) below.

<Card title="MCP server URL" icon="link">
  ```
  https://mcp.blockworks.com/mcp
  ```
</Card>

## Key benefits

<CardGroup cols={2}>
  <Card title="Zero installation" icon="bolt">
    No npm, no npx, no local dependencies or runtime to keep alive. Add a URL to your MCP client and you are done.
  </Card>

  <Card title="OAuth with your API key" icon="lock">
    Browser-based OAuth using your Blockworks API key. Your key is never pasted into a config file on disk.
  </Card>

  <Card title="Always up to date" icon="cloud">
    Hosted infrastructure, so every client picks up new datasets and capabilities without an upgrade step.
  </Card>

  <Card title="Works with any MCP client" icon="plug">
    Any client that supports remote MCP servers over HTTP with OAuth 2.0 can connect to the same URL.
  </Card>
</CardGroup>

## What your assistant can reach

The server exposes Blockworks data as tools your assistant calls on demand. It decides which data to pull from a
natural-language question, then returns structured results it can reason over.

<CardGroup cols={2}>
  <Card title="Market data" icon="chart-line">
    Prices, volumes, and market caps across tens of thousands of assets.
  </Card>

  <Card title="Onchain metrics" icon="link">
    TVL, active addresses, fees, revenue, and transaction activity for major networks and protocols.
  </Card>

  <Card title="Timeseries" icon="chart-area">
    Historical and current metric series with configurable ranges and granularities across a large dataset catalog.
  </Card>

  <Card title="Fundraising" icon="cash">
    Funding rounds, M\&A activity, investors, and company valuations.
  </Card>

  <Card title="Token unlocks" icon="lock-open">
    Vesting schedules, allocation breakdowns, and upcoming unlock events.
  </Card>

  <Card title="News and research" icon="news">
    Aggregated news plus Blockworks research reports, diligence, and quarterly reviews.
  </Card>
</CardGroup>

## The tools

Five tools cover that surface: one that retrieves documents, and two catalog-and-read pairs over the Blockworks
data models.

| Tool                 | Purpose                                                                          |
| -------------------- | -------------------------------------------------------------------------------- |
| `search_documents`   | Search research, news, and curated sources for relevant documents                |
| `tabular_catalog`    | List the tabular models, or describe one in full: columns, filters, sort options |
| `tabular_data`       | Read rows from one tabular model, filtered, sorted, and paginated                |
| `timeseries_catalog` | List the timeseries models, or describe one in full: metrics and granularities   |
| `timeseries_data`    | Read points from one timeseries model over a time window                         |

The dataset tools are catalog-driven: the catalog call returns the exact model slug, field names, and operators,
and the read call accepts nothing that does not match it. That is the same
[discover-then-query](/getting-started/concepts/catalog) model the REST API uses, over the same models.

<Card title="Tools reference" icon="wrench" href="/mcp/tools/overview">
  What each tool does, how they compose, and example phrasing for each.
</Card>

<Note>
  The catalogs list only the models Blockworks publishes to the public API, and which of those your assistant can
  actually read depends on what your API key is entitled to.
</Note>

## How it works

1. You add `https://mcp.blockworks.com/mcp` to your MCP client.
2. The client discovers that the server requires authentication and opens a browser window.
3. You authenticate with your Blockworks API key. The client stores the resulting OAuth credentials, not the key.
4. Your assistant lists the available tools and can start answering questions with Blockworks data.

## Building on MCP

MCP is not only for chat windows on your own machine. The same hosted server is the recommended way to build a chat
experience or an agent product over crypto data: your backend connects to `https://mcp.blockworks.com/mcp` as an MCP
client and hands the Blockworks tools to whatever model you are running.

Anything that speaks MCP server-side can do this, including the Anthropic and OpenAI SDKs, the Vercel AI SDK, and
agent frameworks such as LangChain. Each has its own way of registering a remote MCP server, so follow that
framework's documentation for the wiring; the Blockworks side is the same URL and the same tools described on this
page.

What stays yours in this setup:

* **The model.** Pick any provider and any model, and change it without changing your data layer.
* **The prompts and the orchestration.** System prompts, tool-choice policy, retries, multi-step planning, and
  anything else in your agent loop.
* **The product surface.** Streaming, UI, session state, and how you render tool results and sources.

Blockworks supplies the data layer: document search, dataset queries, and the timeseries catalog and lookups listed
above.

## Requirements

* A Blockworks API key from [app.blockworks.com/account/api](https://app.blockworks.com/account/api).
* An MCP client that supports **remote** servers over HTTP with OAuth 2.0. Clients that only support local `stdio`
  servers cannot connect directly.
* Outbound network access to `https://mcp.blockworks.com`.

## Next steps

<CardGroup cols={3}>
  <Card title="Quickstart" icon="rocket" href="/mcp/quickstart">
    Connect a client and run your first prompt.
  </Card>

  <Card title="Client setup" icon="adjustments" href="/mcp/clients">
    Step-by-step configuration for each MCP client.
  </Card>

  <Card title="What to ask" icon="messages" href="/mcp/prompts">
    Example prompts that play to the server's strengths.
  </Card>
</CardGroup>