git clone https://github.com/eiliyaabedini/aider-mcp.git
   cd aider-mcp

curl -s https://aider.chat/install.sh | sh

wget -qO- https://aider.chat/install.sh | sh

python -m pip install aider-install
aider-install

pip install -e .

pip install -r requirements.txt

# Default model to use if not specified
AIDER_MODEL=gpt-4.1-mini

# Add any other environment variables needed

python tests/test_multiple_ai.py

"AiderMCP": {
      "command": "uv",
      "args": [
        "run",
        "--with",
        "mcp[cli]",
        "mcp",
        "run",
        "/Users/path/to/aider_mcp.py"
      ]
    }

Just ask your MCP client (Claude Desktop) to use the code_with_ai or code_with_multiple_ai tool for coding.
and provide the full path where it should create the files.

Sample of Snake game to run at same time:
Use this directory as WorkDirectory: /Users/eiliya/ai/snake 
This is the prompt for making a Snake Battle Royale,
First create a Readme and Architect file as your plan for implementation and save it as md file then, 
I want you to think deep and design tasks properly, You will use **code_with_multiple_ai **MCP tool to implement the code, so You need to give the coder good context, our coder needs some knowledge as readonly_files, so provide the location of the Readme.md and Architect.md (That you will create and is a big picture of whole project)
Important, this tools let you code in parallel but you need to be careful that your tasks has no dependency, planing is imporant, create different task branches, that each branch can be run without dependency to other branches,
then run tasks of multiple branches at same time,
even if you can design the code the way that all the tasks are seperated and can be run in one attempt (no dependency to each other) then you can run them all in one attempt:
Example:

Branch 1: Front end --> Task1: initiate front end, Task2: implement index page
Branch 2: Back end -->  Task1: initiate backend, Task2: implement api
Branch 3: Database -->  Task1: initiate database, Task2: implement database

Then in each round you can get all the Task 1 to this method
Or 

Branch 1: Task1: Implement index.html, Task2: assemble everything into index.html   
Branch 2: Task1: Implement script.js 
Branch 3: Task1: Implement styles.css 
Branch 4: Task1: Implement script3.js 
Branch 5: Task1: Implement script4.js
Branch 6: Task1: Implement script5.js
Branch 7: Task1: Implement script6.js
Then in each round you can get all the Task 1 to this method, but in task 2 you know all the tasks 1 are implemented, 
then you can have dependency to all other implemented tasks 1 

Game prompt:
"Snake Battle Royale Language: HTML/CSS/JavaScript (vanilla, no frameworks) Libraries: None, pure JavaScript Style: Minimalist design with accent colors in teal and orange Rules: 1. All files should be in the folder snake_battle_royale 1. Keep the code clean and reusable, use functions and classes and split the code into multiple files to keep it organized Guidelines: 2. You can move the snake around the screen using the arrow keys 3. Eat food to grow the snake 4. Avoid hitting the edges of the screen, or the snakes own body, or the enemy snake, or you lose. The game is only over when the players snake is dead. 5. There are at least 3 powerups that change the game mechanics, including effects that show how the powerups work 6. A scoring system that rewards the player for eating food and hitting powerups 7. A simple AI opponent that competes with the player using pathfinding 8. A home screen where the player can hit enter to start the game 9. A game over screen that shows your score and allows you to go back to the home screen or restart the game 10. On the game screen, display a legend containing the powerups and their names, as well as the user and enemy snake's name and score 11. When the game starts, we should have a 3 second countdown where everything is frozen in place, then the game starts"
if **code_with_multiple_ai **tool failed stop the process, I need to check it

python tests/test_multiple_ai.py