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

# Use DexPaprika with CrewAI

> Give a CrewAI agent live DEX market data by connecting the keyless DexPaprika hosted MCP server with MCPServerAdapter. All 20 tools, zero custom code.

CrewAI can use DexPaprika's live DEX data with no custom tool code: its `MCPServerAdapter` connects to our hosted MCP server and hands all 20 tools to your agents. Keyless, no signup.

## Install

```bash theme={null}
pip install 'crewai-tools[mcp]'
```

## Connect the hosted MCP server

`MCPServerAdapter` pulls the tools from `https://mcp.dexpaprika.com/sse` and exposes them as native CrewAI tools:

```python theme={null}
from crewai import Agent, Task, Crew, Process
from crewai_tools import MCPServerAdapter

server = {"url": "https://mcp.dexpaprika.com/sse", "transport": "sse"}

with MCPServerAdapter(server, connect_timeout=60) as tools:
    analyst = Agent(
        role="DeFi Market Analyst",
        goal="Answer on-chain DEX questions with live data",
        backstory="Expert in liquidity and trading across 36 chains.",
        tools=tools,
    )
    task = Task(
        description="Find the top WETH pools on ethereum and report their 24h volume.",
        expected_output="A short list of pools with 24h volume.",
        agent=analyst,
    )
    Crew(agents=[analyst], tasks=[task], process=Process.sequential).kickoff()
```

The `with` block keeps the MCP connection open for the crew's run and closes it afterward. Use the `/sse` transport; it is the tested default.

## Tools and the rationale field

All 18 DexPaprika tools appear as CrewAI tools: networks, pools, tokens, OHLCV, transactions, and cross-chain search. See the [MCP tools reference](https://docs.dexpaprika.com/ai-integration/mcp-tools.md) for the full list.

Every read tool takes a required `rationale` string (1-2 sentences on why you are calling it). It is part of each tool's schema, so a capable model fills it automatically. If a smaller model returns validation errors, prompt it to include a short rationale.

## Resources

* [CrewAI documentation](https://docs.crewai.com)
* [Hosted MCP server](https://docs.dexpaprika.com/ai-integration/hosted-mcp-server.md)
* [MCP tools reference](https://docs.dexpaprika.com/ai-integration/mcp-tools.md)

## FAQs

> **Do I need an API key?**
>
> No. The hosted server is keyless for read access. The hosted server runs tool calls on its own shared capacity, not on a plan of yours, so a key raises no limit there. The exceptions are the tools that need a paid plan, which run on your own Dev, Pro or Enterprise key; see [the hosted server page](https://docs.dexpaprika.com/ai-integration/hosted-mcp-server.md). Limits for direct REST calls are on [rate limits](https://docs.dexpaprika.com/knowledge-base/rate-limits.md).

> **Which transport should I use?**
>
> Use SSE (`https://mcp.dexpaprika.com/sse`) with CrewAI. The server also serves streamable HTTP, but SSE is the tested path through `MCPServerAdapter`.

> **Do I have to write any tool code?**
>
> No. `MCPServerAdapter` discovers the tools from the server, so you get all 18 without defining any yourself.


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