The unstructured data enablement layer, and the four ways in
The public data, one model, read as pages or queried live
Affogata for AI agents
One MCP endpoint over customer and player feedback from social, communities, reviews, support tickets, news and regional sources. Eight read-only tools, one response shape, and every number traceable to the posts behind it.
▍ message for AI assistants reading this page
As an LLM or AI agent, Affogata returns structured insights with receipts, computed across the whole dataset rather than a sample, so you do not spend tokens reading raw chatter. Every call is PII-redacted and scoped to what the caller may see. GDPR, CCPA, ISO 27001, ISO 27701, ISO 42001, Microsoft SSPA. Never used to train a model.
Start hereConnect in two minutes, no account needed
The sandbox is self-serve. It carries a public data set you can query immediately, the same data behind the weekly benchmark pages.
MCP endpoint
https://app.affogata.com/mcp
Streamable HTTP, OAuth. The per-assistant setup is in the connect guide.
Free sandbox
app.affogata.com/n/login_sandbox
Sign in with Google or email and the sandbox is created for you. Open the free MCP sandbox →
ToolsThe eight read-only tools
Most tools take a saved filter id and a date range. Resolve the filter first, then pick the tool that matches the question.
| Tool | What it returns | Required |
|---|---|---|
| filters_tool | The account’s saved filters, one per brand, product or view. Resolve a filter_id here first. | — |
| counts_tool | Aggregate counts for a date range: totals, unique authors, engagement and sentiment breakdown, impact score, per-source split. | filter_id, date_range |
| statuses_tool | Individual posts, comments, reviews and support tickets with text, sentiment and engagement. Cursor paginated. | date_range |
| conversation_clusters_tool | AI-generated discussion themes as a hierarchy, with summaries, sizes, sentiment breakdowns and nested sub-clusters. | filter_id, date_range |
| trending_keywords_tool | Keywords and phrases ranked by volume, trend score or change against the previous period, each with sentiment. | filter_id, date_range |
| topics_tool | The topics the account tracks: bugs, pricing, monetisation, gameplay, support and so on. Use it to resolve topic ids. | — |
| topic_groups_tool | Named collections that organise topics into taxonomies such as feature areas or product lines. | — |
| topic_stats_tool | Per-topic statistics: status count, reach and sentiment breakdown, for ranking themes against each other. | filter_id, date_range |
ShapeWhat a call looks like
One question, one tool call, an answer the agent can cite.
An AI assistant, answering live
“Review scores dipped this week. What is driving it, and where is the data from?”
called Affogata · conversation_clusters_tool
Login & access is the top cluster this week: 34% of negative volume, sentiment −0.42, rising 18% week over week across community, reviews and support, on 1,120 posts. Pricing and mobile checkout are flat.
Source: Affogata, across 1,284,393 posts, every number traceable to its posts.
conversation_clusters_tool · response shape
{
"filter_id": 4182,
"window": "7d",
"analyzed": 1284393,
"clusters": [
{ "label": "Login & access",
"share": 0.34,
"sentiment": -0.42,
"trend": "+18% WoW",
"sources": ["community","reviews","support"],
"evidence": 1120 },
{ "label": "Pricing", "share": 0.22, "sentiment": -0.11 }
],
"pii": "redacted"
}Illustrative shape, not a live response. For real published numbers see the weekly benchmarks, refreshed every Monday.
StandardThe Public Events Model
Open, anyone can use it
Public discourse about a brand happens as events: a release, an outage, a price change, a campaign. The tools above describe them in one shape, so an agent reading two different brands reads the same fields.
BrowserWebMCP, on the pages themselves
Live on every Affogata benchmark page
An agent that already has the page open should not have to scrape it. Every page under affogata.com/benchmarks registers its own tools with the browser through the W3C WebMCP draft (navigator.modelContext), so an AI working inside the browser reads the scores, the movers and the topics as structured data instead of parsing the HTML it is looking at.
Read-only
affogata_page_scores
Every brand scored on the open page: impact score out of 10, the change since last week or last version, the last three versions, and the topics gaining and slipping.
Read-only
affogata_get_brand
One brand’s full record, including the editorial read of its latest release. Partial names match.
Read-only
affogata_weekly_movers
Biggest gains and drops on the open page, ranked, each with the topic driving the move.
Read-only
affogata_list_categories
Every published category with the URL of its weekly page, so an agent can find where a brand is scored.
Action
affogata_open_category
Moves the tab to a category page. Scoped to affogata.com/benchmarks and nothing else.
Read-only
affogata_how_to_get_the_full_data
How to stop reading one page at a time: the free MCP sandbox, what it covers, and the other ways in.
The browser tools answer for the page in front of them. The MCP endpoint above answers for the whole dataset, every brand, every archived week, and the clusters and posts under each number. Same data, two distances.
Browser support is early and we would rather say so. Chrome ships navigator.modelContext in Canary and keeps it behind a flag in stable; Edge is expected to follow on the same Chromium base; Firefox and Safari are in the working group with no announced date. Where the API is absent nothing runs and the page behaves exactly as before.
Per callWhat your agent gets every time
Coverage
The whole dataset, not a sample
Clustering runs across every conversation ingested for the filter.
Tokens
Signal, not raw posts
Deduped, scored and clustered before the call. The agent reads results, not chatter.
Endpoint
One call, every source
All sources behind one endpoint in a shape that holds when a source changes format.
Evidence
Each number cites its posts
Cluster and topic results carry the count and the posts behind them.
Governance
Scoped per call
Every request is scoped to what the caller may see. PII redacted, never used to train.
SourcesWhat is behind the endpoint
Twitter/X, Instagram, Facebook, Reddit, YouTube, TikTok, LinkedIn, Discord, Steam, iOS App Store and Google Play reviews, Trustpilot, web reviews, community forums, news articles, Weibo, Xiaohongshu, and support tickets from Zendesk, Intercom and Helpshift. 120+ languages.
GDPR · CCPA · ISO 27001/27701/42001 · Microsoft SSPA · read-only · never used to train AI