mcp-aider-executor
Exposes local execution of Aider commands as a detached process using MCP.
Documentation
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 Default LLM main model override (i.e. always use this model no matter what)
--edit-format Main model edit format override, regardless of config or LLM inference
--architect Always run in architect mode (--no-architect for never)
--editor-model Default secondary editor model override for architect mode
--editor-edit-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 Only allow operations within the glob pathEnvironment 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.
Frequently asked questions
What is mcp-aider-executor?
mcp-aider-executor is Exposes local execution of Aider commands as a detached process using MCP.
How do I install mcp-aider-executor?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is mcp-aider-executor open source?
Yes — it is hosted on GitHub at https://github.com/RailKill/mcp-aider-executor.
Related MCP tools
Model Context Protocol Servers
The Open-Source Multimodal AI Agent Stack: Connecting Cutting-Edge AI Models and Agent Infra
A MCP for Claude Desktop / Claude Code / Windsurf / Cursor to build n8n workflows for you
MCP server to provide Figma layout information to AI coding agents like Cursor
The world's best AI personal assistant for email. Open source app to help you reach inbox zero fast.
Instant is the best backend for AI-coded apps. You get auth, permissions, storage, presence, and streams — everything you need to ship apps your users will love.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP