---
title: Explore a model
editUrl: https://github.com/Cratis/Screenplay/edit/main/Documentation/screenplay/mcp/explore.md
description: Prompts that ask an assistant what a Screenplay model contains, what depends on what, and where it is incomplete, without changing any file.
---


Everything on this page is read-only. The server answers in compact, paged results,
so the assistant can explore a model of hundreds of files without reading them all.
The examples use the lending library from [Create a model](/screenplay/mcp/create/).

## Get oriented

```text
Use the screenplay server to give me an overview of this application: its
modules and features, how many slices, events and read models it has, and whether
it compiles cleanly.
```

The assistant starts from counts and a `sourceRevision`, then walks down through
modules, features and slices on request. It does not load the whole model.

```text
Walk me through the Lending module: each feature, and for each slice what the
command does and which events it produces.
```

## Find a declaration

```text
Where is BookBorrowed declared? Show me its properties and every file that
contributes to it.
```

```text
Search for anything with "Loan" in its name and tell me what kind of declaration
each match is.
```

Names are case-sensitive, and a module or feature split across files is reported as
one logical declaration with every location listed.

## Follow the connections

```text
Which slices and read models use the BookBorrowed event? Tell me whether each one
produces it or consumes it.
```

```text
What does the BorrowBook command depend on, directly? And what would be affected if
I changed it?
```

The server reports resolved references and the places it cannot see into, such as
code blocks and expressions, so an assistant can say what a rename or removal might
miss rather than assuming nothing else uses a name.

## See how modules and features depend on each other

```text
Show the dependency graph between modules, with the slices and source references
behind each edge. Then show feature cycles and suggest a story order. Do not edit
anything.
```

The assistant calls `dependency-graph` with its default module → module view,
then `view: "cycles", from: "feature", to: "feature"` and `view: "order"`.
For a narrower question:

```text
Which features depend on the Lending.Loans feature? Show incoming dependencies
and their evidence, excluding specifications.
```

Use feature → feature levels, `scope: "Lending.Loans"` and `direction: "incoming"`.
You can also compare mixed levels, such as feature → module, or use `to: "context"`
for imported facts. Follow `nextOffset` with the returned `sourceRevision` as
`expectedSourceRevision`; raise `evidenceLimit` only for edges you need to inspect.

Cycles and suggested order include read-model decisions, not just event flow.
Library's Catalog and Loans form a cycle because Catalog uses loan events and
BorrowBook reads CatalogEntry. This does not add a `PLAY0517` diagnostic. An order
is a suggestion only; ambiguity and unresolved references remain visible. Set
`includeTestOnly: true` when you want specification dependencies too. See the
[dependency-graph reference](/screenplay/mcp/reference/#dependency-graph) for exact options and
coverage.

## Check the specifications

```text
Which slices have no specification that asserts anything? Group them by feature.
```

```text
Show me the specifications that seed a BookAdded event as their starting state,
with the values they use.
```

A specification counts as asserting something when it declares a `then`, including
`then denied`. This finds gaps in the model. It does not run the specifications.

## Read what is on disk

```text
Show me the exact source of the file that declares AddBook, including comments.
```

The assistant reads the original file bytes, so comments and line endings are
preserved. Ask for the merged view when you want one canonical document for the
whole application instead.

## Look for problems

```text
List the diagnostics for the whole application, errors first, with file and line.
```

```text
Are there any problems in just the Catalog feature?
```

An invalid file produces a partial index, never a silently complete one. The
assistant reports parse errors and what could and could not be indexed.

## Keep big questions bounded

On a large model, scope the question before asking for detail:

```text
Without reading every file, tell me which feature has the most slices and show me
only the slices of that feature.
```

Large answers are refused with a smaller alternative rather than truncated. If an
assistant reports that a page was rejected, ask it to narrow the scope or page
through the results.

When you are ready to change something, continue with [Edit a model](/screenplay/mcp/edit/).