llmlingua-cursor

One command. Zero config. LLMLingua-2 prompt compression for every Cursor project.

Installation

npm install -g llmlingua-cursor

Then run the setup wizard:

llmlingua-cursor

You'll see:

  How would you like to install LLMLingua-2?

  [1]  Global  — works across ALL Cursor projects on this machine
  [2]  Project — works only in the current directory

  Enter 1 or 2:

That's it. No manual file editing. No copying paths.

What it does

llmlingua-cursor installs and configures LLMLingua-2 as a local MCP server for Cursor. Every prompt you send gets compressed before hitting the LLM — saving tokens, reducing latency, and keeping your context window lean.

Prerequisites

• Node.js 18+
• Python 3.9+ (must be on your PATH, or a venv at ~/.cursor/venv/)
• Cursor editor

Setup options

Global (recommended)

Configures LLMLingua-2 for every Cursor project on your machine.

llmlingua-cursor
# → enter 1

Files written:

~/.cursor/
├── mcp_llmlingua.py          ← MCP server
├── mcp.json                  ← registers the server with Cursor
└── rules/
    └── llmlingua.mdc         ← Cursor rule (auto-triggers compression)

Per-project

Run from inside your project directory.

cd my-project
llmlingua-cursor
# → enter 2

Files written:

~/.cursor/
└── mcp_llmlingua.py          ← MCP server (always here, one copy)

my-project/.cursor/
├── mcp.json                  ← project-scoped server config
└── rules/
    └── llmlingua.mdc         ← project-scoped Cursor rule

After setup

1. Restart Cursor completely.
2. Open Settings → MCP — confirm a green dot next to llmlingua-compressor.
3. Use Agent mode in Cursor for best results.

Compression runs automatically on every prompt.

How compression works

| Prompt type | Compression rate | |---|---| | Default | 50% | | Explain / summarize | 45% | | Quick Q on large context | 35% | | Logs / stack traces | 40% | | Refactor / rewrite | 55% | | Very large input (>500 lines) | 30% |

Structural tokens ({}, [], def, class, import, etc.) are never dropped — code safety is preserved.

Every compressed response is watermarked:

✦ Optimised by LLMLingua 2

Project structure

llmlingua-cursor/
├── bin/
│   └── cli.js              ← interactive setup wizard
├── scripts/
│   └── mcp_llmlingua.py    ← bundled MCP server (LLMLingua-2 via FastMCP)
├── package.json
├── README.md
└── LICENSE

Troubleshooting

Green dot not showing in Cursor?

• Confirm Python is installed: python --version
• Confirm deps installed: pip install llmlingua "mcp[cli]"
• Confirm ~/.cursor/mcp_llmlingua.py exists
• Fully quit and reopen Cursor

pip install failed during setup?

pip install llmlingua "mcp[cli]"

Tool not triggering automatically? Use Agent mode in Cursor. The .mdc rule instructs the model to call the tool — Agent mode honours this most reliably.

Built by

Mustafa Waqar · Powered by microsoft/llmlingua-2

✦ Optimised by LLMLingua 2