mcpField
M
mcp-aider-executor

mcp-aider-executor

Local
github.com/RailKillDeveloper Tools0Updated 24w ago

About mcp-aider-executor

mcp-aider-executor is a stdio MCP server for LLM agents that launches aider code-editing work in the background. It records prompts and provides process, chat-history, and basic Git status checks for later progress review.

Key features

  1. 01Run aider as a background process
  2. 02Review process status and chat history
  3. 03Check Git status and recent commits
  4. 04Manage branches, stashes, and reversions
  5. 05Restrict operations with path whitelists

Use cases

  1. 01Delegate code edits to aider while an agent continues other work
  2. 02Review progress from an earlier aider run
  3. 03Inspect or prepare a repository before editing

MCP support

Exposes tools for starting aider and checking its progress, prompt, files, configuration, and Git state to MCP clients.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"railkill-mcp-aider-executor": {
"command": "npx",
"args": [
"-y",
"mcp-aider-executor"
]
}
}
}
Run commandnpx -y mcp-aider-executor

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npmmcp-aider-executor1.1.0

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Project Information

stdio MCP server to run aider as a detached background process with basic git tools and status checks. This allows LLMs to start code editing processes freely and check the aider chat history for progress at a later time to allow for agentic swarm behavior.

An .aider.mcp.details.json file is created in the same directory with the process ID and starting datetime. The MCP server provides tools for LLM agents to check if the process is still running, and to check the .aider.chat.history.md for updates.

The original message prompt send via the MCP server to aider is also recorded in .aider.mcp.prompt.txt.

Quick Start

You can run the server directly using npx, passing in environment variables for the endpoint URL and API keys. Provide --whitelist arguments with a permitted glob path otherwise all operations will be denied by default.

{
  "mcpServers": {
    "mcp-aider-executor": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-aider-executor",
        "--whitelist",
        "/home/user/my-project/**",
        "--whitelist",
        "/local/my-application/**"
      ],
      "env": {
        "AIDER_MODEL": "openai/gemma-4-E4B-it-IQ4_XS",
        "OPENAI_API_BASE": "http://127.0.0.1:1337/v1",
        "OPENAI_API_KEY": "your-key-here"
      }
    }
  }
}

Prompting

You can directly tell your LLM to use Aider by providing them with a directory to work in. For example:

Use Aider to create a simple Python hello world console application in "/home/my-project"

For extra Aider options, you should set it properly using environment variables (the env in MCP settings .json), or ask your LLM to create the Aider YAML configuration file for you in the project directory. By default, the server comes with some basic git and file tools to allow the LLM to look around. These tools can be disabled via the options below if you'd like to use a separate MCP server to handle more advanced functions.

Caution

For security purposes, the MCP server does not have the functionality to write API keys into the configuration YAML. Use env instead to provide keys like OPENAI_API_KEY. Don't tell your LLM about it.

Options

You can add these arguments in the npx command.

-h, --help                     Display this message
--model <model>                Default LLM main model override (i.e. always use this model no matter what)
--edit-format <format>         Main model edit format override, regardless of config or LLM inference
--architect                    Always run in architect mode (--no-architect for never)
--editor-model <model>         Default secondary editor model override for architect mode
--editor-edit-format <format>  Editor model's edit format override
--no-add-message-notes         Disables the adding of aider-specific notes to message prompts
--no-register-git-tools        Disables registration of all git tools for this server
--no-allow-git-edits           Disables only the git tools which modify the repo history
--no-register-file-tools       Disables registration of wide file access tools for this server
--whitelist <path>             Only allow operations within the glob path

Environment Prerequisites

The environment running the MCP server must have the following tools installed and available as executable commands in the shell:

  1. node
  2. aider
  3. git

Warning

You should set the number of parallel requests to 2 if you are using the same local endpoint for aider, otherwise your chat session and the background aider process will be stuck waiting for each other.

You should also set the OPENAI_API_KEY environment variable to a non-empty value if you are using a local OpenAI-compatible endpoint because aider requires it even if you don't use an API key.

You still need the openai/ prefix when specifying model names for local endpoints in aider.

Available MCP Tools

These are the list of MCP tools available for your LLM to call.

  • aider_check_git_log: Checks recent commits.
  • aider_check_git_status: Calls git status --short to check the state of the given directory.
  • aider_check_last_prompt: Returns the original prompt message of the last Aider run.
  • aider_check_progress: Checks the Aider chat history and background process status.
  • aider_checkout_git_branch: Switches git branches, creating a new one if it doesn't exist.
  • aider_create_git_stash: Stashes everything to clear the directory for Aider (git stash -u).
  • aider_git_revert: Reverts changes from a git commit by its hash.
  • aider_list_files: Lists all files in a given directory.
  • aider_list_git_branches: Returns a list of git branches in the given directory.
  • aider_mcp_check_whitelist: Checks what glob paths are whitelisted for this Aider MCP server.
  • aider_message_prompt: Starts aider as a background process with a given message prompt.
  • aider_read_config_yaml: Returns the full content of the .aider.conf.yml in the directory.
  • aider_read_file_contents: Reads the contents of a given filepath.
  • aider_setup_config_yaml: Creates the .aider.conf.yml file in the directory.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.