Skip to main content

Using the BrandIndex MCP

Introduction​

The BrandIndex MCP exposes the BrandIndex API to AI agents, so you can query brand health data from your own LLM tools instead of navigating the BrandIndex platform or exporting files. It works as an interface adapter: it translates natural language requests into BrandIndex API calls and returns structured data for your own agent to interpret.

The MCP does anything already possible through the BrandIndex API or platform. It returns aggregated brand metrics. It does not visualise, narrate or export.

What YouGov guarantees​

YouGov guarantees the data the BrandIndex MCP serves. Every value comes from the BrandIndex API and is verified against it.

YouGov cannot guarantee what your own LLM does with that data once it has been returned. Your LLM decides how to read those numbers and what to show you, and models do make mistakes: recalculating a figure, re-basing a percentage, cutting a table short, or claiming a comparison the data does not support.

In practice: treat the retrieved figures as authoritative and your agent's commentary on them as a working draft. Check anything you are about to put in a report or a deck.

See Accuracy and limitations below for the specific failure modes worth knowing about.

Prerequisites​

Before you configure an agent you need:

  • A BrandIndex account with MCP access enabled. MCP access is licensed separately from BrandIndex platform access and from BrandIndex API access. Contact your account manager.
  • At least one previous login to track.brandindex.com. Logging into the platform activates your account, and the MCP cannot authenticate you until that has happened.
  • An MCP-compatible AI client.

Server URL​

https://mcp.brandindex.com/mcp

Opening the server URL in a browser returns "Not Found". This is expected. The server only responds to MCP clients.

Configuring agents​

You authenticate as yourself through YouGov Platform via OAuth 2.0, and the MCP inherits your existing BrandIndex permissions: you can query the regions, sectors and data your account is entitled to.

For setup steps by client, see Client setup.

Using the BrandIndex MCP​

Prompt explicitly. The MCP works best when you name it:

using brandindex, what is Nike's buzz score in the US over the last 3 months

Specify the market​

Every analysis is scoped to a region. If you do not name one, the MCP asks rather than querying every market.

using brandindex, list the sectors available in the UK

Running an analysis​

using brandindex, compare consideration and purchase intent for Adidas and Nike
in the US for the last 6 months

The MCP runs multiple brands as separate queries inside one call, up to 10 queries per call. Each query takes 1 to 4 metrics; ask for more than 4 on one entity and the agent splits it into additional queries.

Applying filters​

using brandindex, show me Vodafone's brand impression in the UK among 18 to 34 year olds

Age filters are fixed three-way bands (18-34, 35-49, 50+ in both the UK and US), not arbitrary ranges. If you ask for a range that doesn't match a band, such as 25 to 34, the agent substitutes the closest band and should tell you it did.

Demographic filters vary by region. If the MCP rejects a filter, ask the agent to list what the region supports.

Filters combine with period comparisons the same way they would in the BrandIndex platform:

using brandindex, for Renault in the UK, compare net satisfaction for the last 30 days
with the 30 days prior, only for people aged 50+
using brandindex, compare Ford's consideration score this quarter versus last quarter
among women in the US

Filtering on a brand metric​

You can also narrow an analysis to respondents who gave a particular answer on one of the brand's metrics, for example current customers, people who would consider the brand, or people who heard positive Buzz. Metric filters are the same in every region.

using brandindex, show me Vodafone's satisfaction in the UK among current Vodafone customers

A metric filter applies to the brand being analysed. You cannot score one brand among another brand's customers or considerers.

Dates​

Dates are absolute, in YYYY-MM-DD format. Your agent converts relative phrases like "last 30 days" into absolute dates before querying. Because of fieldwork lag, a daily series can stop short of the end date you asked for instead of returning a row for every day up to it. The response has no separate field stating that cutoff, so check the date of the last row returned and treat that as the latest available data.

Smoothing and periodicity​

By default, results use BrandIndex's standard 28-day moving average, the same as the platform. Ask for a different window ("a 4-week moving average", "smooth it over a quarter") and the agent translates it to days.

If you want one number per period instead of a smoothed daily trend, ask for periodicity instead: weekly, four-weekly, monthly, quarterly, half-yearly, yearly, or a single whole-period figure.

using brandindex, give me monthly buzz scores for Tesco in the UK for the last 6 months

The two settings are mutually exclusive: a periodicity replaces the moving average rather than adding to it.

Saving an analysis as a chart​

Ask the agent to save an analysis and it persists the same queries as a permanent chart in BrandIndex, then gives you a link to open it in the platform. This is the one thing the MCP does that writes back to your account rather than only reading data - nothing is saved unless you explicitly ask.

using brandindex, save that as a chart

Queries with a periodicity save as a bar chart, matching BrandIndex's own periodicity control; everything else saves as a line chart. Styling is static for now.

Saved charts are created without a title. Rename it in BrandIndex after opening the link if you want one.

What the MCP does not do​

  • It returns aggregated metrics, not respondent-level data.
  • It does not return open-ended verbatims.
  • It cannot rank a sector. You can compare brands you name, but you cannot scan or rank every brand in a category.
  • It cannot base one brand's scores on another brand's audience. Metric filters apply to the brand being analysed, so cross-ownership, overlap and switching questions are out of reach.
  • It does not build dashboards, exports or written commentary - your own agent does that from the data it gets back. The one exception is saving an analysis as a chart (see above), which it does on request.
  • It does not query Profiles or custom trackers, and it cannot read BrandIndex platform assets such as saved analyses or dashboards.
  • It does not run significance testing. Differences between scores, brands or periods are not flagged as statistically significant, and this is not currently available.

Accuracy and limitations​

The introduction sets out what YouGov guarantees. These are the specific failure modes to watch for:

  • Ask your agent to report the numbers it retrieved, not to recalculate them. Recalculation, re-basing and derived percentages are a common source of error.
  • MCP results can differ from the equivalent view in the BrandIndex platform if the underlying query is defined differently, for example a different date window or score type. Same data, different question.
  • A long result set can be truncated by the model rather than returned in full. If a table looks short, ask for the row count.
  • Check the market, metric and date range your agent actually used. It may have resolved an ambiguous request differently from how you intended.
  • A periodicity result can include a trailing period for which fieldwork isn't complete yet. That period comes back with a score of zero rather than being left out or marked incomplete, which can look like a real drop to zero if you don't check it. Before reporting the last period in a periodicity result, confirm it's actually complete.
  • Significance testing is not available. If your agent describes a change as "significant", that is its own judgement and not a tested result.

Troubleshooting​

  • "I'm not seeing a brand or market I expect" - the connector only shows what your BrandIndex account is entitled to. Contact your account administrator to check access.
  • "The numbers don't match the BrandIndex dashboard" - see Accuracy and limitations above.
  • "I can't connect / authorization fails" - confirm you have an active BrandIndex account and have logged into the platform at least once, then retry the connection.
  • Still stuck - see Support below.

Support​

Contact your client service rep or brandindex@yougov.com. To help us investigate, include:

  • The LLM and client you are using (for example, Claude Desktop or ChatGPT, and the model)
  • The exact prompt you entered
  • The exact output you received, as text or a screenshot
  • What you expected to see, and where you checked it, such as the BrandIndex platform view

Reports such as "the numbers are wrong" cannot be investigated without these details.