Configure `Codex` extension in VS Code

[IvanMurzak]

Tasks

Need to implement new Configurator class for Codex in VS Code. To make it use existed Codex configurator as a reference which is located in this folder Unity-MCP-Plugin/Assets/root/Editor/Scripts/UI/AiAgentConfigurators/Impl. And maybe even better to inherit it and to make a new class that must use a different location for the TOML configuration file. Need to find in the official Codex documentation where the VS Code extensions of the Codex saves the TOML configuration file.

• Use existed image of Codex
• Create configurator class in the location: Unity-MCP-Plugin/Assets/root/Editor/Scripts/UI/AiAgentConfigurators/Impl
• Add configurator class in the registry class AiAgentConfiguratorRegistry

[mrwellmann]

@IvanMurzak I have no experience in creating MCPs but I think the /mcp should always be applied no mater on which IDE or CLI Codex runs.

Reasoning:

1. I've been searching through the Open AI docs and found no specific instructions but these examples https://developers.openai.com/codex/mcp/#examples-of-useful-mcp-servers, https://developers.openai.com/codex/integrations/linear/#connect-linear-for-local-tasks-mcp and https://developers.openai.com/resources/docs-mcp/ all have the /mcp.
2. The config.toml can be placed global in ~/.codex/config.toml or project specific in .codex/config.toml (see
   https://developers.openai.com/codex/mcp/?utm_source=chatgpt.com#supported-mcp-features). It is shared between IDE and CLI.

Do you agree I should update CodexConfigurator instead of creating a new one? I would leave the the placement in the project specific config.toml as it is now.

Implementation Result would be:
http://localhost:8080-> http://localhost:8080/mcp
http://localhost:8080/ -> http://localhost:8080/mcp
http://localhost:50888/mcp -> stays unchanged

[IvanMurzak]

@IvanMurzak I have no experience in creating MCPs but I think the /mcp should always be applied no mater on which IDE or CLI Codex runs.

Reasoning:

1. I've been searching through the Open AI docs and found no specific instructions but these examples https://developers.openai.com/codex/mcp/#examples-of-useful-mcp-servers, https://developers.openai.com/codex/integrations/linear/#connect-linear-for-local-tasks-mcp and https://developers.openai.com/resources/docs-mcp/ all have the /mcp.
2. The config.toml can be placed global in ~/.codex/config.toml or project specific in .codex/config.toml (see
   https://developers.openai.com/codex/mcp/?utm_source=chatgpt.com#supported-mcp-features). It is shared between IDE and CLI.

Do you agree I should update CodexConfigurator instead of creating a new one? I would leave the the placement in the project specific config.toml as it is now.

Implementation Result would be:
http://localhost:8080-> http://localhost:8080/mcp
http://localhost:8080/ -> http://localhost:8080/mcp
http://localhost:50888/mcp -> stays unchanged

Hi @mrwellmann , what exactly do you propose to make?

[mrwellmann]

I created a PR with the suggestion #515

[IvanMurzak]

I have no experience in creating MCPs but I think the /mcp should always be applied no mater on which IDE or CLI Codex runs.

Hi @mrwellmann, I did test it on my own in Codex and in Codex in VS Code. They have no difference and both work with existed endpoint. The ending with /mcp is not required.

[IvanMurzak]

Codex in VS Code for a some reason sees only the global .config.toml file which is not how the Codex Configurator is designed to work. All the AI Agent configurators in the project are focused on the local MCP configuration for the local project only, to prevent fighting with other Unity project that also may use AI Game Developer on the same machine.

We may need to create a UI toggle at the AI Agent configurator section for letting user to choose how to configure (local / global) configuration file. And we may add a warning for Codex to use a global config with VS Code.

[mrwellmann]

I tested and agree /mcp does not matter, and I was so sure it was an issue.

On the local .config.toml, it does work for me now with the VS Code extension.
I just confirmed by:

1. I removed the part in the global config.toml and let the Unity extension add it.
2. I tested it via the VS Code extension, and it works.
3. I tested it in another project without the local config, and it did not work.

Which version of the Codex mpc client do you use? There was an update 3 days ago.