Every fix, improvement, and new capability added to the Comscore MCP server since launch.
When an audience like "VP and C-suite" matches several Plan Metrix segments, the assistant now lists them and asks which to use, including combining them, before pulling any data. A clear single match is still used straight away.
Questions that need several lookups at once, such as building an audience from many segments, could be cut off with a "too many requests" error after only a few steps. The usage limit now counts actual data requests and allows twice as many per minute, so these questions run through.
Plan Metrix audiences combining many segments could wrongly come back as "no data". They now return figures; shorter audiences were not affected.
US Digital requests for Adults 18+, Persons 18+, A18+, 25-54 and similar age groups — including audience-overlap questions — now return data instead of an error. If an audience filter genuinely can't be applied, the assistant now tells you rather than showing unfiltered numbers.
Plan Metrix requests return sooner, and follow-up questions about the same sites or categories are quicker still. Other Digital reports also return sooner, and all figures are unchanged.
Comparing competitors over time — for example audience, pages and minutes for four real-estate sites — now comes back in one answer instead of one request per site and metric, so it is much faster. If one name isn't recognised, the others are still reported.
Characterising a brand across five or six audience segments now takes one request instead of one per segment, with each audience shown as its own labelled row. Figures are unchanged.
Cross-platform audience and reach figures were labelled "Total Digital Population" when they are the combined Desktop and Mobile audience; they are now labelled correctly, and the numbers are unchanged. Total Digital Population — which also includes social, and is a larger figure — is now available by asking for it by name.
A connector left idle for more than four hours — any overnight gap — quietly lost its connection. It still showed as "Connected" the next morning but offered no tools, so the only way back was to delete the connector and add it again. A connector in regular use now keeps working, and signing in once a day is no longer required — an active connection is no longer cut off twelve hours after sign-in. Client accounts now stay signed in for up to six months; when a sign-in does need confirming, your assistant asks you to sign in again right in the conversation and then carries on, with nothing to re-add.
Asking for desktop figures on a Multi-Platform subscription returned a subscription error, because desktop-only Media Metrix is a separate product. Those requests now answer from the desktop breakout of the Multi-Platform report, which the subscription does cover, and the response says which report the figures came from — the two use different panels, so the numbers will not match a desktop-only report. Accounts that subscribe to desktop-only Media Metrix are unaffected and keep the same figures as before. Plan Metrix has no desktop breakout to fall back on, so a desktop Plan Metrix request now explains that and names the platforms the subscription does include.
Accounts subscribed to Multi-Platform without a desktop component could not look up a site or brand by name — most requests came back reporting the geography as invalid, even though rankings worked. Name lookups now run against the same platform view as the figures they return. A request for a platform the subscription does not cover now says so plainly instead of reporting an invalid geography.
A connector would report that it needed reconnecting — and offer no tools at all — once its sign-in session lapsed, often within an hour or two of connecting even while it was actively in use. Sessions now stay usable for their full intended lifetime, so a connector keeps working through the working day.
Plan Metrix figures can now be reported site by site for a publisher network — one row per individual site, where before the request came back as an error. Asking for the sites that over-index with an audience narrows the list by index score in the same request.
Audience searches now cover IAB Digital Interests alongside the Plan Metrix survey and demographic segments, and every result says which of those layers it came from. Definitions built on interest signals — venture capital, angel investment, movie genres — can now be reproduced instead of coming back empty.
Every figure now comes with the entity it was measured for — its name, ID and level, so a property is never mistaken for one of its domains — plus the audience segments behind it and the population covered. Where a request matches several entities or age brackets, the alternatives are offered rather than one being chosen silently.
Cross-visiting broken out by individual domains now returns results instead of failing, and unsupported combinations say which sets to use instead.
A Plan Metrix request can now cover several brands and several metrics at once instead of one at a time, so a wide report finishes in a fraction of the requests it used to need — a six-brand, two-metric month went from twelve requests to one. Figures are unchanged; only the number of requests differs.
Our sign-in now follows the OAuth standard more closely, so customers' in-house AI assistants can connect alongside Claude and ChatGPT.
Desktop+Mobile Plan Metrix requests now read Comscore's cross-platform data source — the same one the Plan Metrix UI reports from — instead of combining a desktop and a mobile report. Audience sizes, reach, composition and audience universe figures now match the UI exactly.
If a connection ever stops working, we now show exactly how to fix it instead of a technical error.
Asking for social metrics, posts or rankings by name is more forgiving: one unclear name no longer discards the others in the request, and close matches come back most-relevant-first.
Program and time-based TV viewership results now include a per-demo flag showing whether a figure meets our minimum sample-size threshold, so low-confidence numbers can be called out instead of taken at face value.
Fixed an issue where asking for a broad demographic segment (like "Counties") in a specific TV market could return results for the wrong markets instead of the one you asked about.
A demographic age range that doesn't match a standard bucket (e.g. "children 8-12") now gets pointed to the closest available bucket instead of a plain error. TV markets that are covered by a nearby city's market rather than their own now resolve correctly too.
Hand the assistant a raw station, market, demographic, or network ID — for example from a Snowflake share or a data feed — and it now resolves it back to the name and metadata.
You can now ask for market totals across every Comscore market in one go — for example a cable total per market, by daypart, for a full broadcast month — and get complete coverage of every station that reported. Each response also tells you the scope it covers, so the numbers are ready to use with confidence.
Ask for TV viewership by daypart and you now get a row per daypart — Early Morning, Daytime, Prime and the rest side by side — instead of one blended number. Requests that span hundreds of stations or every Comscore market also complete now: a full month of station-by-daypart viewing across all markets works in a single question, where before it stalled or came back empty. Per-market cable viewing can be requested too, either station by station or as one cable total per market.
You can now ask the assistant questions about your Social audience data, in addition to Digital and TV.
When you ask about a saved group of sites or apps that isn't publicly shared, the assistant now finds it automatically instead of returning no results.
Fixed a login issue that occasionally caused failed requests, added recognition of common network nicknames (e.g. "NBC" for the full network name), and improved handling of time-of-day questions and multi-network requests so results come back more consistently.
When a demographic term you use could match more than one audience segment, the assistant now tells you so you're not left assuming it's an exact match.
Program ranking questions can now be cross-tabbed against a second demographic segment, not just the primary one.
Some demographic breakdowns were sorted by the wrong audience segment and didn't always respect the requested result limit. Also fixed a lookup issue that caused some numeric demographic requests to resolve to the wrong segment.
When a TV question defaults to Live-only viewing, the assistant now tells you so you know DVR and time-shifted viewing weren't included.
Applied a round of security hardening to keep your data and access safe.
TV markets are now consistently labeled "Comscore market" rather than a competitor's trademarked term, near-identical program and series names are kept as separate rows instead of being merged, and the assistant no longer explains these internal formatting rules in its answers.
Corrected an issue that was showing lower-than-actual average audience for Sunday Primetime.
Added clearer names and descriptions so it's easier to see what each capability offers, as part of getting Comscore MCP listed in Anthropic's official connector directory.
If you ask about data outside your subscription, the assistant now gives a clearer explanation of why instead of a generic error.
Improved how the assistant classifies audience segments for cross-visiting and national ad report questions, and tightened rate limits to keep the service responsive for everyone.
Introduced request rate limiting and a round of security fixes to protect the service from abuse.
Added support for many more ways of asking about TV audiences — quarter-hour breakdowns, entity status lookups, multiple demographics and audiences in one question, National Ad Report data, syndicated series rollups, and clearer routing for demographic and industry-ranking questions.
Fixed quarter-hour ratings being incorrectly averaged, gender targeting occasionally matching the wrong audience, and Video Metrix showing blank values for some "% Composition Minutes" questions.
Questions about foot traffic or store visits (not supported by TV measurement) and overly broad demographic lookups now return a clear explanation instead of a confusing result.
Plan Metrix questions can now be broken down by segment crosstab, and mobile app name searches are more accurate.
Audience overlap and duplication questions now handle desktop, mobile, and total-minutes questions more accurately across Digital and Video Metrix.
Fixed an authentication issue affecting some TV questions, and improved the message shown when your account doesn't have TV access.
Improved recognition of station call signs and fixed an issue with the 55+ age range not resolving correctly.
Adjusted how and when your session is refreshed in the background, so you're less likely to be asked to sign in again mid-conversation.
Fixed a login failure for accounts with uppercase letters in their ID, fixed a Video Metrix demographic breakdown issue, and improved product suggestions for ChatGPT users.
You can now ask planning-oriented Plan Metrix questions and Video Metrix questions with platform and demographic breakdowns, in addition to standard Digital and TV data.
Fixed several issues affecting questions asked through Microsoft Copilot, including how filters and content-security settings were handled.
Moved the service to its permanent home at mcp.comscore.com, and added support for frequency-based questions.
Comscore MCP opened up beyond early internal testing — clients with an active Digital or TV subscription can now request access, with clearer messaging for accounts that don't have the right entitlements yet.
Comscore MCP can now be connected as a data source in ChatGPT, in addition to Claude.
Improved how trend-over-time questions are handled and added better support for audience data in India.
Fixed several issues causing incorrect or missing audience figures for markets outside the US, including Latin America and Asia Pacific, and fixed entity names disappearing from some multiplatform reports.
Added internal usage logging and monitoring dashboards to help us spot and resolve issues faster — no change to what you see as a user.
Improved session handling so Copilot users are less likely to be unexpectedly signed out mid-session.
Resolved a series of issues that prevented some questions from working correctly when asked through Microsoft Copilot.
General improvements to how TV-related questions are interpreted and answered.
Comscore MCP can now be connected as a data source in Microsoft Copilot, in addition to Claude.
Improved audience data accuracy for international markets, and the assistant now tells you when a requested month's data hasn't been published yet instead of quietly showing an older month.
Added session refresh tokens so you can stay signed in longer between conversations.
Added easier, more direct ways to ask about audience metrics, reach, and reports, plus the ability to limit how many results come back.
Comscore MCP became available to internal Comscore users first, with default settings for common reports to make getting started easier.
You can now ask questions about Comscore Linear TV audiences, in addition to Digital — the first step toward TV measurement in Comscore MCP.
Added Comscore single sign-on, so you can use your existing Comscore credentials instead of a separate login.
Comscore MCP now works with Claude.ai directly, not just Claude Desktop, and responses use less back-and-forth to reduce wait time.
Added authentication so Claude users can securely connect their Comscore account.
The first version of Comscore MCP goes live, bringing Comscore Digital audience data into conversational AI clients for the first time.