# `Beamlet.MCP.Eval`
[🔗](https://github.com/aaronrussell/beamlet/blob/v0.1.0/lib/beamlet/mcp/eval.ex#L1)

The `eval` tool, which runs Elixir code on your beamlet.

Each use is a fresh evaluation, with nothing carried over from the
last. The code is checked against the token's policy before it
runs. It can call what the policy grants and every module agents
have defined.

The agent gets back what the code printed, then the value of its
last expression. A refused call, an exception or a timeout comes
back as text too, so the agent can see what went wrong and try
again.

A token has the tool when its policy lists `:eval` under `tools`.
`default` does.

## Input

```json
{"code": "Shopping.List.items() |> length()"}
```

* `code` - Required. The Elixir code to evaluate.

## Configuration

    config :beamlet, eval: [timeout: 60_000]

* `:timeout` - How long one evaluation may run, in milliseconds.
  When it runs out, the evaluation stops and the output so far
  comes back. Defaults to 30 seconds. The MCP server's
  `:request_timeout` must stay longer (`Beamlet.MCP.Server`).
* `:max_heap_bytes` - How much memory one evaluation may use.
  Past it, the evaluation stops before it can take the beamlet
  down. Defaults to 128 MB.
* `:max_output` - How much printed output comes back, in bytes.
  Longer output is cut, with a line saying how much was shown.
  Defaults to 32 KB.

# `description`

The description a client lists for the tool, written for the model.

Built when it is read, from the limits above, so it always states
the ones in force.

# `get_schema`

# `mcp_schema`

# `mcp_schema!`

# `mcp_schema_changeset`

---

*Consult [api-reference.md](api-reference.md) for complete listing*