Problem

The dv-connect MCP configuration reference (references/mcp-configuration.md) marks per-tool branch boundaries with **If TOOL_TYPE is \x`:**` (bold) in sections 0, 1, 2, 5, and 8.

This is the same markdown construct used for within-block step labels in those sections — e.g. **Generate a unique server name**, **Update the configuration file:**, **Important notes:**. So branch boundaries are not structurally distinct from the sub-steps inside a branch.

An agent parsing the file has no reliable signal for where one tool's instructions end and the next tool's begin, which risks instruction bleed across tools (e.g. applying a Cursor step during a Codex flow).

Proposed fix

Promote each per-tool branch to a real ### subsection so the boundary sits structurally above the **bold** sub-labels:

## 5. Register the MCP server

Execute only the subsection matching your `TOOL_TYPE`; skip the others.

### 5a. Copilot
...
### 5b. Claude
...
### 5c. Cursor
...
### 5d. Codex
...

• Apply uniformly to every per-tool conditional section (1, 2, 5, 8).
• Add an explicit "execute only the subsection matching your TOOL_TYPE" routing line at the top of each.
• This is a Level-3 reference file (no token budget), so the added headings are free.

Context

Surfaced during PR #84 review. Deferred to its own issue to keep that PR scoped to Codex enablement; this refactor touches all four tool flows and is best reviewed independently.