CLI - MCP Tool Limits?

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor CLI

Describe the Bug

The CLI does not share the same MCP tool limits as the agent window or IDE. There is a stricter cap on available tools allowed.

“”“”
Error: Provider Error
Too many MCP tools are enabled for this
model. Please disable some MCP servers and
try again.
“”“”

Steps to Reproduce

Anytime MCPs are loaded into a CLI session.

Expected Behavior

The CLI should enjoy the same MCP tool capacity as the Cursor app (agent window/IDE).

Operating System

MacOS

Version Information

Version: 3.11.19
VS Code Extension API: 1.125.0
Commit: bf249e6efb5b097f23d7e21d7283429f0760b740
Date: 2026-07-12T21:39:24.175Z
Layout: Agent Window
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Darwin arm64 25.5.0

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

5

Hey, thanks for the detailed report. I dug into what’s going on.

The Too many MCP tools are enabled for this model error isn’t a separate CLI limit. It’s a per-model provider cap on the number of tools. On the Cursor side, the MCP tools limit is shared across the CLI and the IDE, but the CLI currently sends all configured MCP tool definitions in a single request, while the IDE effectively sends a lot fewer. Because of that, the CLI hits the model’s tool limit sooner, which makes it feel like there’s a stricter cap.

For now, the workaround is:

  • Disable or trim some MCP servers so there are fewer active tools in your CLI session.
  • Or switch to a model with a higher tool cap.

We’re aware of the CLI eagerly loading all tool definitions and we’ve logged an issue for it. I can’t share an ETA yet, but I’ll post an update in the thread when I have one.

7

Thanks for explaining. This indeed can be quite a nuisance as I setup my CLI configuration to match the MCP tooling and environment that exists in the Cursor app.

I am using the Auto model and have no other alternative.

Could I work around this by loading a small subset and progressively loading the remainder once the agent session is initialized?

8

Progressive loading isn’t supported right now. The CLI ships all configured MCP tool definitions at the start of the session, and you can’t load or switch servers at runtime. /mcp only supports list and list-tools, so there’s no runtime toggle. The “lazy” tool loading you’re asking about is exactly what we’ve logged internally.

Since you’re on Auto and don’t want to pin a specific model, the only practical lever is to reduce how many MCP servers are active in a given CLI session, so the total tool count doesn’t hit the per-model cap:

  • Keep a lightweight per-project set of servers via .cursor/mcp.json instead of loading the full global set.
  • Check the actual tool count with cursor-agent mcp list and cursor-agent mcp list-tools <server> to see what’s eating the limit.
  • Leave enabled only the servers you really need for the current task.

Once there’s an update on lazy-loading, I’ll reply in the thread.