Ultra feedback: Auto is extremely slow and MCP tools are unreliable after recent updates

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

  • Models, pricing & API keys — Auto model
  • MCP & tools
  • Editor, Tab & Chat — Agent

Describe the Bug

I’m an Ultra subscriber and a heavy daily user of Cursor. Since the recent Cursor updates, agent workflows have regressed significantly in both performance and reliability.

Auto has become extremely slow during normal agent work, especially while discovering and invoking tools. In addition, multiple configured MCP tools fail intermittently, and some connected/authenticated MCP servers or tools are not exposed to the agent at all. Tool visibility can differ between sessions: a tool may be available in one session and missing or unusable in another.

This is not limited to one isolated MCP server. The combined Auto latency and inconsistent MCP discovery/invocation make multi-step agent workflows unreliable.

Steps to Reproduce

  1. Open an existing project in Cursor with multiple MCP servers configured and authenticated in Cursor Settings.
  2. Start a new Agent chat and select Auto.
  3. Ask the agent to perform a multi-step task that requires discovering and invoking one or more configured MCP tools.
  4. Observe that the agent/tool orchestration is significantly slower than before the recent updates.
  5. Observe that some MCP calls fail with errors, or configured tools are not available in the agent’s tool catalog at all.
  6. Start another Agent session or reload Cursor and repeat the same task.
  7. Observe that MCP tool visibility/availability may differ between sessions.

The failures are intermittent, but the latency regression is frequent.

Expected Behavior

  • Auto should provide responsive and predictable routing/execution for an Ultra user.
  • Configured and authenticated MCP servers and tools should be consistently discoverable by the agent in every session.
  • MCP calls should execute reliably.
  • When a tool is hidden or rejected, Cursor should provide an actionable reason instead of a generic error or silent disappearance.

Screenshots / Screen Recordings

Not attached yet. Sanitized MCP logs can be provided with a focused reproduction.

Operating System

  • macOS 25.5.0, arm64

Version Information

IDE:

  • Version: 3.13.25
  • Commit: 31e8d61c448c7472e371505838a0fe34083dad50
  • Architecture: arm64

For AI issues: which model did you use?

  • Model: Auto
  • Plan: Ultra

Request ID

A valid Request ID is not available for this broad/intermittent report yet. I have not copied the example ID from the template because it does not belong to my session. I can reproduce the issue with privacy disabled and add the relevant Request IDs and sanitized MCP logs.

Additional Information

The issue became materially worse after recent Cursor releases. Please investigate:

  • Auto routing and latency for Ultra users
  • MCP catalog synchronization and session-to-session consistency
  • MCP tool discovery/exposure to the agent
  • Recent regressions in MCP invocation
  • Better diagnostics for tools that are configured but hidden, rejected, or failing

Does this stop you from using Cursor?

  • Sometimes — I can sometimes use Cursor, but MCP-heavy agent workflows are frequently too slow or unreliable to complete normally.
2

Hi there!

We detected that this may be a bug report, so we’ve moved your post to the Bug Reports category.

To help us investigate and fix this faster, could you edit your original post to include the details from the template below?

Bug Report Template - Click to expand

Where does the bug appear (feature/product)?

  • Editor, Tab & Chat (autocomplete, Composer, in-editor agent)
  • Terminal & commands
  • Models, pricing & API keys (availability, Auto/Max, BYOK/Bedrock)
  • MCP & tools
  • Cloud Agents & Automations (cursor.com/agents, scheduled/event)
  • BugBot & Code Review
  • Cursor CLI
  • Cursor Mobile
  • Remote (SSH / Dev Containers / WSL)
  • Account, billing & login
  • Something else…

Describe the Bug
A clear and concise description of what the bug is.

Steps to Reproduce
How can you reproduce this bug? We have a much better chance at fixing issues if we can reproduce them!

  1. …
  2. …
  3. …

Expected Behavior
What is meant to happen here that isn’t working correctly?

Screenshots / Screen Recordings
If applicable, attach images or videos (.jpg, .png, .gif, .mp4, .mov)

Operating System

  • Windows 10/11
  • MacOS
  • Linux

Version Information

  • For Cursor IDE: Menu → About Cursor → Copy
  • For Cursor CLI: Run agent about in your terminal
IDE:
Version: 2.xx.x
VSCode Version: 1.105.1
Commit: ......

CLI:
CLI Version 2026.01.17-d239e66

For AI issues: which model did you use?
Model name (e.g., Sonnet 4, Tab…)

For AI issues: add Request ID with privacy disabled
Request ID: f9a7046a-279b-47e5-ab48-6e8dc12daba1
For Background Agent issues, also post the ID: bc-…

Additional Information
Add any other context about the problem here.

Does this stop you from using Cursor?

  • Yes - Cursor is unusable
  • Sometimes - I can sometimes use Cursor
  • No - Cursor works, but with this issue

The more details you provide, the easier it is for us to reproduce and fix the issue. Thanks!

7

Hey @Semih_ERSOZ

Auto slow: Auto picks a model per request and we’ve been tuning that routing lately, so it’s leaning on heavier (slower) models for some requests. If you want consistent speed, just pin a specific fast model in the Chat dropdown (a Composer model is a good default) instead of Auto.

MCP “not found” between sessions: this one’s on us, an MCP reliability issue we’re tracking, not your setup. Quick fixes while we work on it:

  • When a tool drops mid-chat, start a fresh chat or reload the window (the catalog goes stale in long chats).
  • If you’ve also got a hand-defined atlassian server in .cursor/mcp.json alongside the Atlassian plugin, remove the manual one, they collide.
  • Toggle the server off/on in Settings > Tools & MCP to force a reconnect.

Next time it drops, could you grab a Request ID (three-dot menu → Copy Request ID) plus the MCP logs (Output panel → MCP: atlassian)? With privacy disabled that’ll let me trace exactly why it dropped.