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

# Tool detail

> One tool in full: outcomes, latency distribution, retries, response size and what it gets called alongside.

`/m/:id/tools/:name` is one tool, with the same metrics as the overview scoped down to it.

## What is on the page

|                          |                                            |
| ------------------------ | ------------------------------------------ |
| **Calls**                | In range                                   |
| **First-call success**   | With retries beside it                     |
| **Average response**     | In bytes and in tokens                     |
| **p95 latency**          | The bucket the 95th-slowest call falls in  |
| **Cost**                 | This tool's response bytes, priced         |
| **Schema size**          | What it costs every session, called or not |
| **Outcome breakdown**    | ok, empty, bad\_args, tool\_error, crashed |
| **Latency distribution** | The four buckets                           |
| **Calls per day**        | With ok and empties per day                |
| **Follows**              | Pairs involving this tool                  |

The outcome breakdown and latency distribution render from the same components and the same shapes as the [overview](/metrics/overview), so the two screens cannot disagree about what "ok" counts.

## Calls per day carries three series

Calls, `ok` and `empties` per day. Reading them together is usually how a problem gets located in time: an empties line that lifts on a particular day points at a deploy, a data change, or an upstream that started returning nothing.

Gap-filled, so a quiet day plots as zero.

## Follows

Up to ten [tool pairs](/metrics/tool-pairs) where this tool is either side. Nightly, so it lags by a day.

Pairs are the fastest way to understand *when* a tool is used rather than how often. A tool that always follows one particular search is part of that search's job, whatever its description says.

## Registered but never called

A tool that appears in the [startup payload](/sdk/what-is-sent) but has no recorded calls still has a page. Every figure reads zero, `never_called` is set, and the schema size is the one real number on it — which is the point. See [Dead tools](/tools/dead).

A name that is neither registered nor recorded returns `404`.

## The name is what the client sent

The recorded `tool_name` is `request.params.name` — what the client actually asked for, not what you meant to register. If a tool is missing from the list but you know it runs, check the name matches exactly.

## Related

* [Tool health](/tools/health)
* [Outcomes](/metrics/outcomes)
* [Latency](/metrics/latency)