Aerith Admin MCP Server

The Aerith Admin MCP (Model, Controller, Presenter) server implements a MANUS-inspired development workflow for RBAC dashboard applications with browser automation capabilities. This server is designed to be run locally and accessed by Cursor IDE's MCP integration.

Overview

This server provides a structured approach to development through a 5-step workflow:

USER_INSTRUCTION - Define development tasks with clear goals
TASK_PLANNING - Break down tasks into specific subtasks
INFORMATION_GATHERING - Collect relevant information from various sources
ANALYSIS_AND_ORCHESTRATION - Analyze information and create execution plans
RESULT_SYNTHESIS - Execute steps and generate comprehensive reports

Installation

# Clone the repository
git clone https://github.com/your-org/aerith-admin.git
cd aerith-admin/mcp

# Run the installation script (creates virtual environment and installs dependencies)
./bin/install.sh

# Activate the virtual environment
source bin/activate_venv.sh

Usage

The server can run in two modes:

HTTP Mode (Default)

python server.py --port 8090

This starts the server on port 8090 (or specify a different port). The server provides a REST API and Server-Sent Events (SSE) for real-time updates.

STDIO Mode

python server.py --stdio

This mode is designed for integration with other tools, communicating through standard input/output using JSON-RPC protocol.

Resilient Mode

For production or extended development sessions, you can run the server in resilient mode, which automatically monitors the server health and restarts it if it crashes:

# Using the convenient script
./bin/start_cursor_server.sh --resilient

# Or directly using the resilient server script
./bin/run_resilient_server.sh --mode http --port 8090

The resilient mode includes:

Continuous health monitoring
Automatic restart on crashes
Graceful shutdown handling
Heartbeat detection and logging

Monitoring

You can manually check server health or restart it using the monitoring script:

# Check if the server is running and healthy
./bin/monitor_server.py --check-only

# Force restart the server
./bin/monitor_server.py --force-restart

Cursor IDE Integration

This MCP server is specifically designed to work with Cursor IDE. Cursor can connect to the server to utilize its capabilities directly from the editor.

Setup Cursor Integration:

Make sure the MCP server is running in HTTP mode: python server.py --port 8090

Cursor automatically detects the MCP server using the .cursor/mcp.json configuration:

{
  "mcpServers": {
    "aerith-admin-mcp": {
      "url": "http://localhost:8090/sse"
    }
  }
}

Open the Aerith Admin project in Cursor IDE
Use the Cursor MCP integration UI to interact with the server

Project Structure

mcp/
├── bin/                    # Executable scripts
│   ├── activate_venv.sh    # Script to activate virtual environment
│   ├── install.sh          # Installation script
│   ├── check_env.py        # Environment validation script
│   └── run_tests.py        # Test runner script
├── requirements.txt        # Production dependencies
├── requirements-dev.txt    # Development dependencies
├── server.py               # Main MCP server implementation
├── tests/                  # Test suite
│   ├── conftest.py         # Pytest configuration and fixtures
│   ├── README.md           # Testing documentation
│   ├── test_browser_automation.py # Browser automation tests
│   ├── test_core_workflow.py      # Workflow step tests
│   ├── test_integration.py        # End-to-end integration tests
│   ├── test_resources.py          # Resource access tests
│   ├── test_server_modes.py       # Server operation mode tests
│   └── test_utils.py              # Utility function tests
└── venv/                   # Virtual environment (created by install.sh)

Development Setup

This project uses a dedicated virtual environment for development:

# Run the installation script
./bin/install.sh

# Or manually set up the environment
python -m venv venv
source bin/activate_venv.sh
pip install -r requirements-dev.txt

# For browser automation testing
python -m playwright install

Testing

Tests are written using pytest and located in the tests/ directory.

Running Tests

Use the provided script to run tests:

# Run all tests except browser and slow tests
./bin/run_tests.py -v

# Run with coverage report
./bin/run_tests.py --coverage

# Include browser automation tests
./bin/run_tests.py --browser

# Include slow integration tests
./bin/run_tests.py --slow

# Run specific test files or patterns
./bin/run_tests.py test_core_workflow

Environment Variables

MCP_DEBUG=true - Enable debug logging (set automatically by activate_venv.sh)
Additional environment variables can be configured as needed

API Documentation

Tools

The server provides the following tools:

Instruction Management

create_instruction(title, description, goal, priority) - Create a new development instruction
get_instruction(instruction_id) - Retrieve an existing instruction
build_feature(title, description, goal, priority) - High-level orchestration to build a complete feature

Workflow Steps

create_task_plan(instruction_id, subtasks) - Break down an instruction into specific subtasks
gather_information(instruction_id, sources) - Gather information from various sources
analyze_and_orchestrate(instruction_id, analysis, execution_plan) - Analyze and create an execution plan
execute_step(instruction_id, step_id, execution_details) - Execute a specific step in the plan
generate_final_report(instruction_id, include_details) - Generate a final report

Browser Automation

run_browser_agent(goal) - Run a browser-use agent to achieve a specified goal

Filesystem Tools

tree_directory(directory_path, max_depth, show_files, show_hidden, pattern, exclude_common, custom_excludes) - Generate a tree representation of a directory structure similar to the Unix 'tree' command

Git Tools

git_status(detailed) - Show the working tree status
git_log(count, show_stats, path, author, since, until) - Show commit logs
git_diff(file_path, staged, commit, compare_with) - Show changes between commits or working tree
git_branch(create, delete, remote, branch_name, base_branch) - List, create, or delete branches
git_checkout(branch_name, create, force) - Switch branches or restore working tree files
git_commit(message, all_changes, amend) - Record changes to the repository
git_push(remote, branch, force, tags) - Update remote refs along with associated objects
git_pull(remote, branch, rebase) - Fetch from and integrate with another repository
git_add(paths) - Add file contents to the staging area

Resources

The server provides these resources:

file://{path} - Get file contents by path
project://structure - Get the project structure as a dictionary
instructions://list - Get list of all instructions

Data Storage

All instructions and related data are stored in JSON files in the .aerith/instructions directory.

Logging

Logs are stored in .aerith/logs/mcp_server.log and also output to stderr. When MCP_DEBUG=true is set (default in the development environment), detailed debug logging is enabled.

License

MIT License

Aerith 管理仪表板