# Read Agent tool calls

> See which customer-domain MCP tools agents called, separate tool use from setup handshakes, and check delivery when data is missing.

- Section: Guides
- Updated: 2026-10-07
- Canonical: https://knitknot.ai/docs/read-agent-tool-calls/
- Publisher: KnitKnot, the AI competitive positioning platform (https://knitknot.ai)

---

## At a glance

1. **Choose a window**: Open Traffic → Agent tool calls and select seven, 30, or 90 days.
2. **Read tool use**: Compare estimated calls for each read-only tool with the daily chart.
3. **Check the basis**: Keep protocol handshakes separate and inspect Delivery if telemetry is unavailable.

If your workspace has Agent delivery, open [Traffic → Agent tool calls](https://app.knitknot.ai/traffic/tools). This view counts use of the read-only MCP tools on your customer domain. It is separate from the [authenticated KnitKnot workspace MCP server](/docs/connect-to-ai-tools/), which an assistant reaches after signing in to your account.

## Read the usage

Choose **7d**, **30d**, or **90d** above the chart. The page starts at 30 days. **Estimated tool calls** counts tool requests in that window; **Tools used** counts distinct tools with recorded calls. The totals are sample-weighted estimates, shown with an approximation mark. Do not present them as exact request logs.

The **Tools called** table names each tool and its share of the estimated tool-call total. `get_page` retrieves one published Agent-version page; `search_content` searches the domain's published content index; `get_facts` reads approved facts and their evidence receipts. Sort the table to see which tool was used most.

The daily chart appears when at least two dates were measured. One measured day is a total, not a trend; unmeasured dates stay blank. Choose a wider window as data arrives rather than reading a blank date as zero.

## Separate connection activity from content use

**Protocol handshakes** are client setup calls. They appear in their own summary and, when recorded, below the table by method. They are excluded from the tool-call total. A handshake does not mean the client retrieved a page or fact.

A tool call shows that an agent requested information from the customer-domain endpoint. It does not show that the agent used the response in a final answer. To check what an AI engine cited or said, inspect a later [benchmark](/docs/read-your-report/); to check page fetches, use [Agent activity](/docs/read-ai-traffic/).

## When there is no data

**No agent tool calls** can mean the endpoint is available but no tool was called in the selected window. **Data unavailable** means the telemetry for this deployment could not be read; follow **View delivery status** to [Content → Delivery](/docs/check-agent-delivery/). If the page says tool calls are not enabled for the workspace, the feature is not available there. Check [Content → Interfaces](/docs/manage-agent-interfaces/) for the customer-domain MCP interface's effective state.