Tools

The FullVision MCP mirrors the API: reports, people, audiences and custom reports, plus insights and data-health tools. 38 tools in total. There is no raw SQL and no schema exploration — you ask for a report, not a query. See Setup to connect.

Use this page when you want to know what your AI agent can actually reach, or when it calls the wrong tool and you want to name the right one for it.

Reports

These answer "where does the money come from?" — by page, channel, search keyword, paid campaign, email campaign or form.

One tool per report. Each has a group form (omit the addressing key → ranked list) and a single form (pass the key → one entity's deep-dive). fields trims the response blocks; it never widens one.

Amounts are integer cents in each row's currency. Never sum across currencies.

People

These answer "who?" rather than "how much?". People and journeys return personal data — names, email addresses, event timelines — so an agent should quote them, not dump them.

Audiences

Build a segment as a criteria tree, check it, then activate it. The loop is build → preview → revise → commit, then attach a destination.

The criteria grammar is documented at Criteria DSL; the create_audience and preview_audience tool descriptions carry the same grammar inline, so your agent has it without reading the docs.

Custom reports

Publish live charts on a FullVision-hosted custom report your team (or a client) can open, the way you would publish an artifact. A custom report has an id, a title and a URL on app.fullvision.io, and can be pinned as a tab in the app sidebar for everyone in the workspace. Add, edit, remove or reorder its blocks, move a block to another custom report, or create a separate custom report for a separate audience. An agency working across clients uses each client's own key, so every custom report lives in that client's workspace. The same custom reports are available over the REST API.

A block is a { chart, size } card. Its chart is a live request (source: "report"): a report tool call pinned with its params, re-run on every view rather than a copy of the numbers. Data charts (source: "data") are refused on custom reports. Cards with the chart fields at the top level are still accepted until 2026-12-31. Pass a rolling range (last_30_days, 90d, last_year...) for a custom report that stays current, or fixed from/to for a period that should not move. Each block takes a viz. Any viz works on any report: the viz picks which part of the answer it draws.

• Over time, from the report's series (need params.granularity and a metric): line, area, and bar with a granularity. Up to four more metrics in columns add one line or bar each (for example spend and revenue on one chart). combo draws metric as bars and columns[0] as a line on its own axis. stacked_bar and stacked_area also need params.series_by and draw one layer per segment.
• From the ranked rows (need a metric): bar without a granularity, bar_table (labelled horizontal bars), pie and donut (the largest slices, the rest folded into Other), and scatter (also needs metric_x for the x axis). table shows up to six columns of the ranked rows.
• From the totals (need params.totals: true): big_number (one metric), big_number_sparkline (also needs params.granularity), and funnel (two to six total columns, one per step, in funnel order).
• heatmap shades a grid. With a metric, columns names the row and column dimensions, e.g. visitors with block: "active_hours", metric: "page_views", columns: ["day_of_week", "hour"]. Without one, the first column labels each row and the rest are its cells, e.g. block: "retention", columns: ["cohort_period", "m0", "m1", "m2"].

The row vizes (bar without a granularity, bar_table, pie, donut, scatter, table, heatmap) can draw another list from the answer with block: visitors has devices, browsers, countries, active_hours and retention; channel-report has sources; ad-report has landing_pages. Block rows are summed per label and sorted by the metric.

An invalid combination is rejected with a message naming the field to add.

A key can only request a report it is allowed to read, and a custom report holds at most 24 blocks. people and journeys cannot be requested, because they return individual people.

Viewer filters

A custom report, and any single chart on it, can carry filters: controls that anyone viewing it uses to change the dates or narrow the data without editing the saved custom report. Add them when the user says which filters a custom report should offer.

account_id and exclude_campaign need options ({ value, label }, 1 to 50): read the account ids and campaign ids from ad_report first. A custom report holds at most 6 filters and a chart at most 3. Omit default to keep each chart's saved value until the viewer picks one.

{
  "title": "Paid acquisition",
  "filters": [
    { "key": "date_range", "default": { "range": "last_30_days" } },
    { "key": "platform" },
    { "key": "exclude_campaign", "options": [{ "value": "21063556273", "label": "Brand" }] }
  ],
  "blocks": [
    {
      "size": "wide",
      "chart": { "source": "report", "report": "ad-report", "title": "Spend", "viz": "line", "metric": "spend", "params": { "range": "last_30_days", "granularity": "daily" } },
      "filters": [{ "key": "campaign_type" }]
    }
  ]
}

For each chart, its own filter wins over the custom report's filter, which wins over the chart's saved params. A custom-report filter must apply to at least one chart on it, and a chart filter to its own chart; otherwise the write is refused, and removing the last chart a custom-report filter applies to is refused until the same call removes the filter. A chart a custom-report filter does not apply to says so ("Not filtered by Platform"). On update_custom_report, filters replaces the custom report's list whole ([] removes them); on a card edit, filters replaces that card's list.

Insights

Fixes FullVision found for this workspace (landing-page conversion fixes, SEO, ads, retention), each with the issue, the fix, the page URL and an estimated monthly earning. An agent can apply a fix in your codebase or CMS and record it. An applied fix with a metric becomes an experiment, so the agent can also read whether it worked.

Data health

Setup

Five tools let a coding agent install FullVision in your project. They appear only on a connection you approved in the browser as a workspace owner, never for a pasted API key. See Install FullVision with your coding agent.

Setup steps, in order: account, stripe_connect, tracker:marketing, tracker:app, events, checkout, channels. Each has a status: pending, waiting_on_user, live, skipped or blocked. A pending step that belongs to another project carries the reason other_repository.