Tools & resources
Beta
The tool set is part of the MCP preview and may change before general availability.
These are the tools the assistant calls on your behalf once connected — you don't call them directly; you ask questions in plain language and the assistant picks the right ones. Every tool is read-only and scoped to your account's reporting data.
Start here
The assistant is told to orient itself before answering, using two tools:
| Tool | What it does |
|---|---|
describe_reports | Describes what your account can report on — the KPI catalog (built-in, custom and AI), whether AI analytics is on, your AI call-purpose vocabulary, and the available report views. |
find_entities | Resolves names to the IDs the filters need — a queue, hunt group, user, phone, extension or number. (Filters take IDs, not names.) |
Analytics
| Tool | What it does |
|---|---|
analytics_summary | Aggregated call KPIs over a date range, broken down by a view (by time, user, group, number, and combinations), with intervals (hour/day/week/month/…) and time bucketing (hour-of-day, day-of-week, …). |
analytics_intelligence | AI Call Insights — breakdowns by call purpose, sentiment, customer-satisfaction and agent-performance scores. Requires AI analytics; it errors clearly if that's off, so an empty answer is never mistaken for "zero." |
Individual calls
| Tool | What it does |
|---|---|
query_calls | Lists individual call records — one row per call, per ring leg, or per queue/hunt-group session. Supports filters, sort and paging, and reports the full match total separately from what it returned. |
get_call | One call's full journey by id — every leg in order (who rang and answered, ring and talk time, transfers), queues entered with wait times, and the AI summary and scores when present. Reports whether audio exists, but never returns a recording URL. |
get_call_transcript | The full speaker-separated transcript plus AI summary, bullet points, purpose, sentiment and scores for one call. Requires AI transcription. |
Search
| Tool | What it does |
|---|---|
search_calls | Finds transcribed calls by what was said — a full-text search over transcripts and summaries — combined with AI filters (sentiment, action-required, urgent-only). Requires AI. |
Saved reports
| Tool | What it does |
|---|---|
list_saved_reports | Lists your account's own saved report definitions — their view, interval, range, filters and schedule. |
run_saved_report | Runs one saved report by id, using its stored view, KPIs, filters, date range and time zone. |
Resources
Alongside the tools, the server offers two short guides the assistant reads to interpret your data correctly — for example, so it doesn't read a missing AI score as a zero:
| Resource | What it is |
|---|---|
telebroad://reports/semantics | How Telebroad call reports work — the data model (calls, legs, parties, queue sessions), how each KPI is defined and where they overlap, that durations are in seconds, and how time bucketing and filters behave. Required reading before interpreting any KPI. |
telebroad://reports/recipes | Which tool answers which question — worked recipes for common questions, plus the mistakes that produce plausible-but-wrong answers. |
Two things that shape every answer
- Time zone. Durations are always in seconds, and timestamps are UTC. For
anything grouped by day or hour, the assistant passes an IANA time zone (e.g.
America/New_York) so day boundaries land where you expect — ask in your own zone and say so if the assistant seems to be using UTC. - AI features. Several tools (
analytics_intelligence,get_call_transcript,search_calls) only work if AI analytics/transcription is enabled on your account. They fail with a clear message rather than returning empty, so you can tell "no AI" apart from "no results."describe_reportstells you whether AI is on.