cursor-cli 0.1.2

A wrapper for cursor-agent with formatted output support

Cursor CLI

A Python wrapper for cursor-agent with enhanced output formatting support.

Installation

From PyPI

pip install cursor-cli

From source

cd cursor_cli
pip install -e .

Or run directly as a module

python -m cursor_cli [args...]

Features

• Real-time Output Streaming: Outputs cursor-agent results in real-time as they arrive (default mode)
• Formatted Stream-JSON Output: The output is parsed and formatted for better readability
• Color-coded Output: Different message types (system, user, thinking, assistant, tool_call) are displayed in different colors
• Aggregated Output: Messages of the same type are aggregated together instead of being displayed on separate lines
• Extended Permissions Setup: --danger flag for setting up cursor-agent permissions

Usage

Quick Start (推荐)

# 默认流式输出模式
cursor-cli "Analyze this project"

# 等效于:
cursor-cli --output-format stream-json --stream-partial-output -p "Analyze this project"

Text Mode

# 使用 --text 切换到文本输出模式
cursor-cli --text "Analyze this project"

# 等效于:
cursor-cli --output-format text -p "Analyze this project"

Danger Mode (Extended Permissions)

Setup extended permissions for cursor-agent in .cursor/cli-config.json:

# 在用户 home 目录下创建 ~/.cursor/cli-config.json
cursor-cli --danger

# 在指定目录下创建 .cursor/cli-config.json
cursor-cli --danger /path/to/project

This will create/update the config file with these permissions:

{
  "permissions": {
    "allow": [
      "Shell(*)",
      "Read(*)",
      "Write(**/agents/**/*)",
      "Write(**/.agents/**/*)"
    ],
    "deny": []
  }
}

Runner-specific Options

# 默认流式模式（推荐）
cursor-cli "Your prompt here"

# 文本输出模式
cursor-cli --text "Your prompt here"

# 禁用颜色
cursor-cli --no-color "Your prompt"

# 禁用格式化输出（原始 JSON）
cursor-cli --no-format "Your prompt"

# Show runner help
cursor-cli --runner-help

Programmatic Usage

from cursor_cli import CursorCLIRunner, cursor_cli

# Simple function call
result = cursor_cli("Hello, what can you do?")
print(result)

# Streaming output
for line in cursor_cli("Explain Python", stream=True):
    print(line)

# JSON output
result = cursor_cli("Analyze this", json=True)
print(result)

# Using the runner class
runner = CursorCLIRunner(use_colors=True)
exit_code = runner.run(["-p", "your prompt", "--output-format", "stream-json", "--stream-partial-output"])

Using the Formatter Directly

from cursor_cli import StreamJsonFormatter
import sys

# Create formatter
formatter = StreamJsonFormatter(output=sys.stdout, use_colors=True)

# Process JSON lines
formatter.process_line('{"type":"system","subtype":"init","model":"Composer 1"}')
formatter.process_line('{"type":"user","message":{"role":"user","content":[{"type":"text","text":"Hello"}]}}')

# Finalize when done
formatter.finalize()

Output Format

When formatting is enabled, the output is displayed as:

[SYSTEM:init] Model: Composer 1 | CWD: /path/to/dir | Permission: default | Session: abc12345...

[USER] your prompt text here

[THINKING] (thinking content if any)

[ASSISTANT] Response from the assistant...

[TOOL_CALL:started] 🔧 Tool #1: $ command to execute

[TOOL_CALL:completed]    ✓ Success

[TOOL_CALL:started] 📝 Tool #2: Creating analysis.txt

[TOOL_CALL:completed]    ✓ Created 15 lines (1234 bytes)

[RESULT] 🎯 Completed in 2.5s | 📊 Stats: 2 tools, 350 chars generated

Color Scheme

• SYSTEM: Cyan
• USER: Green
• THINKING: Dim gray
• ASSISTANT: Yellow
• TOOL_CALL: Magenta
• RESULT: Bold Green

CLI Options

Option	Description
"prompt"	Default streaming mode with formatted output
--text "prompt"	Text output mode
--danger [path]	Setup extended permissions (default: ~/.cursor)
--no-color	Disable colored output
--no-format	Disable output formatting (raw JSON)
--runner-help	Show help message

API Reference

cursor_cli() Function

cursor_cli(
    prompt: str,
    model: str = "composer-1",
    stream: bool = False,
    json: bool = False,
    **extra_args
) -> Union[str, dict, Iterator[str]]

参数	类型	默认值	说明
prompt	str	必填	发送给 cursor-agent 的提示
model	str	"composer-1"	使用的模型
stream	bool	False	是否使用流式输出
json	bool	False	是否返回 JSON 格式
**extra_args	-	-	额外的命令行参数

返回值:

• stream=False, json=False: 返回 str 文本输出
• stream=False, json=True: 返回 dict 解析后的 JSON
• stream=True: 返回 Iterator[str] 生成器，逐行输出 JSON

Message Types

The formatter handles the following message types from cursor-agent's stream-json output:

Type	Description
system	System initialization and configuration
user	User input/prompt
thinking	AI thinking process (streaming deltas)
assistant	AI response
tool_call	Tool invocations (shell commands, file operations, etc.)
result	Task completion with duration and statistics

Supported Tool Types

Tool	Icon	Description
shellToolCall	🔧	Shell command execution
writeToolCall	📝	File creation
readToolCall	📖	File reading
editToolCall	✏️	File editing
listToolCall	📂	Directory listing
searchToolCall	🔍	Code search

Development

Release to PyPI

# Install build dependencies
pip install build twine

# Release to PyPI (will prompt for token)
python scripts/release.py

# Release to TestPyPI first
python scripts/release.py --test

Requirements

• Python >= 3.8
• cursor-agent must be installed and available in PATH

License

MIT