Native LSP support or first-class project-scoped LSP via MCP would significantly improve Codex for multi-language microservices

What feature would you like to see?

I would like Codex to support one of these paths as a first-class capability:

1. Native LSP support in Codex, or
2. First-class project-scoped LSP via MCP

For my workflow, this would materially improve coding quality and debugging speed.

Why this matters

I use Codex heavily for coding and debugging across a multi-language microservices stack:

• Java
• Go
• Python
• TypeScript / JavaScript
• Shell
• YAML / JSON
• Jenkins pipelines

In these codebases, text search and file reading are useful, but there is a meaningful gap without semantic language-server support.

LSP-style capabilities would help with:

• go-to-definition
• find references
• symbol search
• cross-file rename safety
• diagnostics / errors / warnings
• better understanding of interfaces, implementations, generated code, and large service boundaries

This is especially valuable in:

• large repos
• multi-module Java codebases
• Go services with interface-heavy design
• TypeScript monorepos
• intense debugging across service boundaries

Current limitation

Today, Codex feels much weaker than an IDE-integrated coding agent when it comes to semantic understanding of large codebases.

The most practical workaround right now seems to be:

• rely on IDE LSP/indexing outside Codex, or
• expose LSP servers through MCP manually

But that is not a first-class experience.

Best acceptable outcome

I would be happy with either of these:

Option A: Native LSP support

Codex can directly use language servers for supported languages.

Option B: First-class project-scoped LSP via MCP

Codex makes it easy to connect project-local language servers through MCP, with project/session scope rather than only global scope.

This feels especially important because project-scoped LSP servers are often the right implementation path in practice.

Alternative / related direction

A strong near-term alternative would be:

• make project-scoped MCP a first-class feature specifically for LSP-backed tools
• document recommended LSP-via-MCP patterns for Java, Go, Python, and TypeScript
• make Codex treat those MCP tools as a first-class code-understanding surface

Related prior discussions

I found these related issues:

• #3023 Project scope MCP
• #647 Any Plans or Interest in File Relationship Analysis?

This request is narrower and more explicit: semantic code understanding through LSP, either natively or through first-class project-scoped MCP support.

Environment

• Codex CLI version: 0.114.0
• Platform: macOS
• Use case: multi-language microservices, heavy debugging, deep repo navigation

Additional information

I am specifically asking because this is one of the biggest differences in day-to-day coding quality between Codex and IDE-integrated agents that benefit from stronger semantic editor context.