> ## Documentation Index
> Fetch the complete documentation index at: https://docs.attention.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Analysis

> Run natural-language analysis across one or many calls in a single request.

`ask_attention` is the workhorse of multi-call analysis. Instead of fetching each transcript and asking the LLM to reason over them, hand a list of call IDs (or a deal ID) plus a prompt to Attention's analysis engine — it processes all the calls server-side and returns a focused answer. This is dramatically more efficient than calling `get_call_details` in a loop.

## Tools

### ask\_attention

<Note>Scope: `mcp:read`.</Note>

Analyze calls by running AI analysis across **up to 25 calls** in a single request. Provide either a list of call IDs (typically returned from [`search_calls`](/mcp/tools/calls#search_calls)) or a deal ID — the engine pulls all conversations associated with that deal automatically.

<ParamField query="prompt" type="string" required>
  The question or analysis prompt to run against the calls. Be specific about what to surface (objections, pricing mentions, next steps, etc).
</ParamField>

<ParamField query="conversation_ids" type="array">
  List of call IDs to analyze. Provide this **or** `deal_id`.
</ParamField>

<ParamField query="deal_id" type="string">
  Deal ID to analyze all associated calls. Provide this **or** `conversation_ids`.
</ParamField>

**Returns:** A Markdown answer to the prompt, grounded in the supplied calls. Includes references back to specific calls when relevant.

**Example prompt:** *"Across these 12 calls with Datadog, what are the recurring objections and how have they evolved over the deal's lifetime?"*

<Tip>
  The recommended pattern is **`search_calls` → `ask_attention`** rather than fetching each transcript with `get_call_details`. The engine reads transcripts, scorecard results, and intelligence items in a single pass.
</Tip>