Using the Crunch MCP

Introduction

MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems.

The Crunch MCP exposes the Crunch API to AI agents enabling users to use their own LLM-driven tool-calling environments for querying and analyzing Crunch data. MCP works as an interface adapter focused on optimizing user workflows.

Configuring Agents

All MCP clients require the following URL as part of their setup instructions.

https://<your-org-subdomain>.crunch.io/mcp/

For YouGov customers, the subdomain would be "yougov"

MCP Clients fall in to two categories - strict and custom. Strict clients require each user to authenticate individually. An example of a strict client is Claude Desktop. Other clients support using an API key and custom configuration - this is useful for service accounts.

The Crunch MCP supports both use cases. See below for the use case that matches your need.

Using the Claude Desktop App

For the latest Claude setup instructions, please refer to Claude Documentation for details on how to setup connectors in the Claude Desktop App. The Crunch MCP can be added as a custom connector using the URL above. Your Claude plan will dictate how you configure the MCP.

Crunch is working on registering the MCP with the Claude Connector market place to simplify setup - Watch this space!

If you are unable to setup the desktop app, contact support@crunch.io

Custom Configurations (Claude CLI, VSCode, Cursor etc)

Custom Configurations require an API key. Contact support@crunch.io or your account manager to request a key with MCP permissions.

After you have received your API key, configure your agent with URL above and the custom header "X-Crunch-Api-Token".

Claude

Using the CLI

claude mcp add --transport http crunch \
  https://<your-org-subdomain>.crunch.io/mcp/ \
  --header "X-Crunch-Api-Token: <your-api-key>"

Manually Edit Configuration

An example entry in your Claude configuration file would look like:

      ...
      "mcpServers": {
          "crunch": {
              "type": "http",
              "url": "https://<your-org-subdomain>.crunch.io/mcp/",
              "headers": {
                "X-Crunch-Api-Token": "<your-api-key>"
              }
          }
      },
      ...

Reference: Claude MCP Documentation

Using Crunch MCP

Once you have the MCP configured in your agent, check it is successfully connected (e.g. using /mcp in Claude).

To explore datasets in Crunch, you must ask the LLM to operate on the specific dataset URL or ID. For example, you might want to list demographic variables in dataset test-dataset.

using crunch, list demographic variables in dataset https://<your-org-subdomain>.crunch.io/dataset/ab12cdef

You can ask Crunch to create data tables e.g.

using crunch, show me tv show awareness segmented by gender

The MCP will request you provide the dataset URL if you do not provide it as this example shows.

I'll need your Crunch dataset URL. It should look something like:

  https://<subdomain>.crunch.io/dataset/<dataset_id>/

Have fun exploring your data with natural language!

Help Center