Composer 1.5 fails to detect and run MCP tools

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Composer 1.5 frequently fails to detect and run MCP tools. It tried to access resources using normal web-based URLs and fails with messages that say “URL requires login, cannot access”. MCP servers give access to such internal URLs.

Steps to Reproduce

Install several MCP servers, and ask composer to specifically do something that requires MCP access.

Expected Behavior

All models in cursor, including Composer 1.5 are able to use MCP tools effectively without mistakes.

Operating System

Windows 10/11

Version Information

Version: 2.5.25 (user setup)
VSCode Version: 1.105.1
Commit: 7150844152b426ed50d2b68dd6b33b5c5beb73c0
Date: 2026-02-24T07:17:49.417Z
Build Type: Stable
Release Track: Early Access
Electron: 39.4.0
Chromium: 142.0.7444.265
Node.js: 22.22.0
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.19045

For AI issues: which model did you use?

Composer 1.5

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

1 Like

Hey, thanks for the report. This is a known limitation in Composer 1.5 when working with MCP tools. Instead of calling MCP tools, it tries to access resources directly via URL. A similar bug has already been reported here: Composer 1.5 cannot see full response from Figma official MCP

For now, the workaround is to use Claude Sonnet 4.6, Opus 4.6, or GPT 5.3 Codex for tasks that require MCP. These models call MCP tools correctly.

A few questions to help us debug:

  1. Which MCP servers do you have configured?
  2. Do other models (Claude, GPT) work with the same MCP servers without issues?
  3. Can you share the Request ID from the chat where Composer 1.5 didn’t use MCP? Click the three dots in the top right of the chat, then Copy Request ID.

The team is aware of the issue. Your report helps us prioritize it.

Yup. This is what I figured out after trial and error. Thanks for confirming :slight_smile:

  1. Internally developed.
  2. Yup.
  3. I don’t have that. I just reuse the same chat prompt by resubmitting old prompt with new model (why keep stale conversation in the history). But I keep observing this problem. Specially this get really annoying in ‘Auto’ mode where cursor tries to use composer and keeps failing. Looking at thinking trace and other behaviors we can infer which model the auto is trying to use. I think this should be fixed with high priority because cursor advertises and recommends people to use ‘Auto’ to control costs.
1 Like

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.