---
title: "Code mode"
description: "Several calls in one round trip."
canonical: "https://atelier201.com/docs/mcp/code-mode"
updated: "2026-11-06"
---

# Code mode

`https://mcp.atelier201.com/mcp/code` exposes two tools, `search` and
`execute`, instead of one per capability.

`search` runs a model-written filter over the OpenAPI document and returns the
operations that matched. `execute` runs a model-written script whose
`request()` goes through the SAME route matching, scope check and handler an
HTTP call does. There is no second authorization surface.

**Why it exists:** a change that needs five calls costs five round trips and five
tool-result payloads in the model's context. Code mode makes it one.

**When not to use it:** if the agent makes one call, the standard tools are
simpler and their descriptions are better.
