Before we start: This thing can completely destroy your system, files , projects and even worse... so be careful. By default this thing has permission to do whatever it wants on your computer.
IMPORTANT NOTE: This project is originally based on the wonderwhy-er/DesktopCommanderMCP project. It started as a telemetry-free version and has been transformed into a full fork, the main difference is that it is telemetry free and the license will always be MIT. no new features or anything, actually we will probably be behind the main version and we may strip some features down to keep the bare minimum and avoid tool bloating. but no telmetry. no license change.
DevControlMCP is an MCP (Model Context Protocol) tool that enables Claude desktop app to execute terminal commands and interact with your file system. It turns Claude into a powerful assistant for coding, system management, and file operations.
- Terminal Operations: Execute commands with output streaming, timeouts, and background execution
- Process Management: List and manage running processes
- File Operations: Read/write files, create/list directories, move files, and get metadata
- Advanced Search: Find files by name or search within file contents using ripgrep
- Code Editing: Make surgical text replacements or full file rewrites
- URL Content: Fetch and process content from URLs
- Claude Code Integration: Delegate complex tasks to Claude Code CLI instances via the
claude_codemeta-tool - Line-based File Reading: Read files with line offset and limits for better handling of large files
- Audit Logging: Track all tool calls with automatic log rotation
- Fuzzy Search Logging: Comprehensive logging and troubleshooting for search operations
- Node.js v18 or higher
- Claude Desktop app with Pro subscription
- Claude CLI installed globally (for
claude_codetool) - install withnpm run install:claude-cli
git clone https://github.com/regismesquita/DevControlMCP.git
cd DevControlMCP
npm run setupThe claude_code tool requires the Claude CLI to be installed and configured:
# Install Claude CLI
npm run install:claude-cli
# One-time permission acceptance (required)
claude --dangerously-skip-permissions
# Follow the prompts to accept permissionsNote: The --dangerously-skip-permissions flag is required for the claude_code tool to function without interactive prompts.
The following npm scripts are available for working with fuzzy search logs:
# View recent fuzzy search logs with detailed formatting
npm run logs:view
# View more logs by specifying a count
npm run logs:view -- --count 20
# Analyze logs with statistics and recommendations
npm run logs:analyze
# Export logs to JSON or CSV format
npm run logs:export
# Clear logs (with confirmation prompt)
npm run logs:clear
# Clear logs without prompt
npm run logs:clear -- --forceAll tool calls are automatically logged to ~/.devcontrol-mcp/tool-calls.log for auditing purposes. The log file is automatically rotated when it reaches 10MB in size. The system retains the 5 most recent log files and automatically removes older logs to prevent unbounded disk usage.
Tool call logs include:
- Timestamp
- Tool name
- Arguments passed to the tool (when applicable)
| Category | Tool | Description |
|---|---|---|
| Configuration | get_config |
Get the complete server configuration |
set_config_value |
Set a specific configuration value | |
| Terminal | execute_command |
Run a terminal command |
read_output |
Read output from a running session | |
force_terminate |
Stop a running terminal session | |
list_sessions |
List all active terminal sessions | |
list_processes |
List all running processes | |
kill_process |
Terminate a process by PID | |
| Filesystem | read_file |
Read local files or URLs with line offset and limits |
read_multiple_files |
Read multiple files at once | |
write_file |
Write to a file (replace contents) | |
create_directory |
Create a new directory | |
list_directory |
List files/directories | |
move_file |
Move or rename files | |
search_files |
Find files by name | |
search_code |
Search for patterns in file contents | |
get_file_info |
Get file metadata | |
| Text Editing | edit_block |
Make surgical text replacements |
| Meta-Tool | claude_code |
Execute prompts via Claude Code CLI with full capabilities |
filepath.ext
<<<<<<< SEARCH
content to find
=======
new content
>>>>>>> REPLACE
The claude_code tool allows you to delegate complex tasks directly to Claude Code CLI instances. This is particularly useful for:
- Complex multi-step coding tasks
- Advanced Git operations
- Terminal command sequences
- Web research and summarization
- Tasks requiring specialized Claude Code capabilities
{
"prompt": "Refactor this function to use async/await",
"workFolder": "/path/to/project",
"tools": ["Bash", "Read", "Write", "Edit"]
}prompt(required): Natural language instruction for Claude CodeworkFolder(optional): Absolute path to working directorytools(optional): Array of Claude tools to enable (e.g., ["Bash", "Read", "Write"])
Configure Claude CLI paths in your DevControlMCP config:
{
"claudeCliPath": "/custom/path/to/claude",
"claudeCliName": "claude-custom"
}claude_code tool bypasses DevControlMCP's internal permission system (allowedDirectories, blockedCommands) because it delegates to an external Claude CLI process. The Claude CLI operates with its own (skipped) permissions.
The following configuration options can be set using the set_config_value tool:
| Option | Description | Default |
|---|---|---|
allowedDirectories |
Directories that can be accessed (empty array for full access) | [] |
blockedCommands |
Commands that cannot be executed | Various system commands |
defaultShell |
Shell to use for command execution | bash (Unix) or powershell.exe (Windows) |
fileReadLineLimit |
Maximum number of lines to read from a file | 1000 |
fileWriteLineLimit |
Maximum number of lines to write to a file | 50 |
maxLineCountLimit |
Maximum line count for file reading (prevents memory issues on very large files) | 1000000 |
binaryFileSizeLimit |
Maximum size for binary files in bytes | 10485760 (10MB) |
claudeCliPath |
Absolute path to Claude CLI executable | undefined |
claudeCliName |
Name of Claude CLI binary | 'claude' |
You can customize the descriptions of any tool by setting environment variables. This allows you to tailor tool descriptions to your specific needs without modifying the code.
The environment variable pattern is:
MCP_DESC_<TOOL_NAME>="Your custom tool description"
Where <TOOL_NAME> is the uppercase name of the tool with any non-alphanumeric characters replaced by underscores.
Examples:
# Custom description for get_config tool
export MCP_DESC_GET_CONFIG="View the current configuration settings for the DevControlMCP server"
# Custom description for execute_command tool
export MCP_DESC_EXECUTE_COMMAND="Run a command in the terminal, with output streaming and timeout support"
# Custom description for read_file tool with more specific details for your environment
export MCP_DESC_READ_FILE="Read a file from your project directory or fetch content from a URL"These environment variables can be set in your shell profile for persistence or right before launching the server.
- Set
allowedDirectoriesto control filesystem access for most tools - Be cautious when running terminal commands as they have full access to your system
- Use a separate chat for configuration changes
claude_codetool: Bypasses internal permission controls and delegates to Claude CLI with--dangerously-skip-permissions- Monitor the audit logs regularly to track tool usage
- Set appropriate limits for binary file size and line reading to prevent memory exhaustion
- This tool can completely destroy your system, files, projects and even worse... so be careful. By default, DevControlMCP tools have broad permissions. The
claude_codetool, in particular, operates with full system access by design, bypassing DevControlMCP's specific permission settings.
This release includes several improvements from the upstream project:
- Line-based File Reading: Files are now read line by line instead of character by character, with configurable limits
- Streaming File Reading: Using readline for memory-efficient processing of large files
- Binary File Protection: Size limits for binary files to prevent memory exhaustion
- Audit Logging: All tool calls are now logged with timestamps and arguments
- Log Retention Policy: Automatic cleanup of old log files to prevent unbounded disk usage
- Fuzzy Search Logging: Comprehensive logging for edit operations with similarity scores and execution times
- Logging Utilities: New npm scripts for viewing, analyzing, exporting, and clearing logs
- Levenshtein Distance: Added fastest-levenshtein library for improved string comparison
- Enhanced Configuration Options: New options for controlling line limits, binary file size limits, and maximum line counts
- Customizable Tool Descriptions: Easily change tool descriptions using environment variables with length validation
- Claude Code Integration: New meta-tool for delegating complex tasks to Claude Code CLI instances
All features have been implemented without telemetry, maintaining our commitment to privacy.
This project is licensed under the MIT License.
Original work by Eduard Ruzga.
Modifications (Mostly removal of stuff) starting from commit 3bdaa965b4a77f64a9f8b751680bd1d90e651851 by RDSM.