Model Context Protocol (MCP) is an open protocol that standardizes how applications provide context to LLMs. It creates a common standard for LLMs to consume resources (e.g. CSV files), define and call tools and prompts.
Arato as an MCP client
Arato can act as an MCP client, allowing you to import resources and tools defined by your MCP server for use in your experiments.
Importing Datasets
Open the DataHub and click the Import from MCP server button. Your server should expose an SSE transport
Enter an MCP server URL and and optionally set an API key for your server. We currently support Bearer tokens
A list of available resources will appear. Only resources with a text/csv mime type are shown
Select a resource to preview the dataset on the right-hand side
Click Save to create the dataset
The new dataset will appear in the DataHub and can be used as any other dataset in Arato
Importing Tools
Open the model configuration settings
Turn on the Add Tools option
In the MCP section, enter your MCP server URL and optionally an API
Click Import
The tool definitions from your server will be populated in the Tools Schema section. You can manually adjust the schema as needed.
Refer to the Understanding Tools section for more information on using tools in Arato
MCP server requirements
Your server should expose the SSE transport
Your server can optionally expect an Bearer token for authentication
Your server should expose the tools and resources capabilities
Only resources with text/csv mime type are supported
Arato as an MCP server
Arato exposes your datasets and prompts to MCP clients, such as Claude Desktop.
You can connect your Claude Desktop (or other MCP clients) to Arato by using our local MCP Server. As MCP is an evolving standard, this is currently the recommended approach. Once remote remote MCP servers are more widely supported, we will also serve a remote MCP server option.
Installing the Arato MCP server
Follow your MCP client’s documentation on installing MCP servers.
For example, here is how to set it up on Claude Desktop
Add the settings below in your Claude Desktop MCP settings
Generate an API key for your Arato account here
Copy the API key and replace the API_KEY value in the MCP settings
Restart Claude Desktop for the changes to take effect
Your prompts and dataset will now be available in your Claude Desktop
{
"mcpServers": {
"arato": {
"command": "npx",
"args": [
"arato-mcp-server"
],
"env": {
"ARATO_SERVER": "https://app.arato.ai",
"API_KEY": "ar-..."
}
}
}
}