Ask AI and MCP

Ask AI

The Ask AI tab answers questions about the contact center in plain words, for example:

• Which queue had the worst service level last week?
• Where do callers hang up in the IVR?
• Who are the agents with the longest handle time this month?

How an answer is built

1. The model reads the question and plans up to three structured queries against the metrics catalog: which metrics, groupings, filters and period. It knows the real queue and agent names, so "the billing queue" becomes a filter on that queue.
2. The queries run like any report, with the asking user's permissions. Agents get answers from their own rows only.
3. A second model call writes the answer from the numbers that came back.

The model never writes code or database queries and never sees data beyond the query results. Each answer shows the queries it ran and their numbers, so it can be checked, and each query can be opened in Explore to continue from there.

Save as report under an answer opens its first query in Explore, ready to save as a saved report. The period stays relative: an answer about yesterday becomes a report that always shows the day before, not the date you asked on.

New report with AI

To build a report that you keep, choose New report with AI above Saved reports in the Reports tab and describe it in your own words, for example:

• Every week: service level and abandoned calls per queue, day by day
• Agents ranked by average handle time last month
• Calls by hour of the day over the last 30 days

The AI picks the metrics, groupings, filters, time breakdown, chart and sort from the metrics catalog, and a relative period that fits: a weekly report gets Last week, a daily one Yesterday. It also suggests a name and a description. The report opens in Explore with a preview on live data, and nothing is saved yet. Check the numbers, change anything you like, then choose Save: the name and description are filled in for you. From then on it is an ordinary saved report: it can be shared and emailed on a schedule.

In Explore, Change with AI edits the open report from a sentence such as split it by agent or only the Sales queue. Whatever you don't mention stays as it is. Changing a saved report this way changes nothing until you save it.

Before the report is shown, it is test-run with your permissions. If it doesn't run, for example because a grouping does not apply to a metric, the AI gets the reason and corrects the report once. When no metric fits the description, it says what is missing instead of guessing. Agents can only build reports from the metrics they can see.

Setup

Ask AI needs AI to be enabled for your environment; we set up the AI model provider with you. Without it the tab says AI is not configured.

MCP server

Floor Walker has an MCP (Model Context Protocol) server at POST https://<your-floor-walker-host>/analytics/mcp, so AI assistants such as Claude can explore the reports directly.

• Transport: Streamable HTTP, stateless, JSON responses.
• Authentication: like the reporting API: an API key with the mcp scope in X-Api-Key, or a Flex user's token. A key without the mcp scope gets 403 forbidden.
• Tools per key: the mcp scope gives the aggregate tools, list_metrics, query_metrics and ask. The contact tools, search_contacts and get_contact, show customer numbers, AI analysis and recordings, so a key also needs the contacts:read scope for them; without it they are not listed. A Flex user's token gets every tool.
• Permissions: the caller's. With an agent's Flex token, the tools return that agent's rows only.

Tool	What it does
list_metrics	The catalog: metrics with definitions and units, dimensions, and which sources they apply to.
query_metrics	Runs a structured query: metrics grouped by dimensions over a period, optionally by time bucket. Durations are seconds, percentages 0–100.
ask	The same as the Ask AI tab: a question in plain words, answered with the queries it used.
search_contacts	Finds contacts, like the Contacts tab. A key needs contacts:read as well.
get_contact	One contact end to end: IVR steps, queue and agent legs, AI analysis. A key needs contacts:read as well.

Connect Claude Code

claude mcp add --transport http floorwalker-reports https://<your-floor-walker-host>/analytics/mcp \
  --header "X-Api-Key: <your-api-key>"

Ask us for a key with the mcp scope for each assistant, and add contacts:read only if the assistant should see individual contacts. An API key sees every row. For a user-level assistant, pass that user's Flex token instead, so their permissions apply.