Desktop harness lags really badly on token-high dependant MCP servers

Resolved 💬 2 comments Opened Jun 2, 2026 by GameRipper665 Closed Jul 15, 2026

What version of the Codex App are you using (From “About Codex” dialog)?

26.527.60818

What subscription do you have?

Plus

What platform is your computer?

x64

What issue are you seeing?

The Codex desktop app starts lagging really badly when I use the Roblox Studio MCP, it usually happens when Codex searches through a lot of Roblox scripts, greps stuff, lists scripts, or opens/reads many scripts from the project. After that the whole app gets slow / barely usable. Typing has delay, scrolling is laggy, and clicking around in the main thread window is delayed too.

Roblox Studio itself is still usable, so I think the problem is more with the Codex desktop app handling lots of MCP output at once, this also makes Codex take up an absurd amount of memory.

What steps can reproduce the bug?

  1. Open Codex desktop app.
  2. Connect the Roblox Studio MCP server.
  3. Open a Roblox project that has a lot of scripts (doesn't have to be a lot, just a lot of action and searching has to be going on).
  4. Ask Codex to search through the project or find code in many scripts.
  5. Let it call the MCP tools a bunch of times, like searching, grepping, listing scripts, and getting script contents.

After this the Codex app starts getting very laggy and slow to use.

What is the expected behavior?

The app should stay usable even if the MCP returns a lot of text or script results, maybe big MCP outputs could be collapsed, shortened, or rendered in a lighter way (like in the Antigravity IDE) so the whole desktop app does not start lagging, as i don't see this issue appearing in the CLI version of Codex.

Additional information

I mostly noticed this with Roblox Studio MCP because it can return a lot of scripts and search results really fast, this might (likely) happen with other MCP servers too if they return a lot of output.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗