Supported coding agents

Reference for the five coding agents AI coding setup supports, including each agent's context file, terminal support, and options.

EConvaiMcpAgent lists five coding agents that Set Up AI Coding (MCP) can configure: Claude Code, Cursor, VS Code, Gemini, and Codex. Use this reference to find the exact context file, terminal behavior, and agent-specific options for each one.

Agents and context files

Agent	Context file	Location	Terminal configuration	CLI command
Claude Code	CLAUDE.md	Project root	Available	claude
Cursor	.cursor/rules/convai.mdc	.cursor/rules/ directory	Not available (GUI app)	—
VS Code	.github/copilot-instructions.md	.github/ directory	Not available (GUI app)	—
Gemini	GEMINI.md	Project root	Available	gemini
Codex	AGENTS.md	Project root	Available; also has the auto-approve option below	codex

Claude Code, Codex, and Gemini are CLI agents — the setup dialog's terminal option is only enabled for these three, and it writes the exact claude, codex, or gemini command as the terminal's startup line. Cursor and VS Code are GUI apps, so the terminal option is disabled when either is selected.

How the primer is written to the context file

When you enable Add Convai instructions to the agent's context file, the plugin copies the Convai primer from Resources/AgentPrimer/ConvaiAgentPrimer.md into the agent's context file, wrapped between two sentinel comments:

<!-- >>> Convai primer (managed by Convai; edits between sentinels are overwritten) -->
...primer content...
<!-- <<< Convai primer -->

Re-running setup with the same agent finds the existing sentinel-delimited block and replaces only what is between the sentinels, so content you add outside the block is preserved. If the context file does not yet contain a Convai block, the write mode you choose in the dialog decides what happens:

Write mode	Behavior	Default
Append	Adds the sentinel-delimited block to the end of the file, preserving any existing content.	Yes
Replace	Overwrites the file with only the sentinel-delimited block.	No

Codex auto-approve

Codex has one option no other agent has: Let Codex run unattended with no approval prompts (bypasses Codex's sandbox), off by default. Enabling it changes the terminal launch line from codex to codex --dangerously-bypass-approvals-and-sandbox, so Codex executes every command with no approval prompt and no sandbox. This option only appears when Codex is the selected agent, and it has no effect unless the terminal option is also enabled.

Next steps

AI coding assistantAI coding assistant quick startConvai AgentSkills