AEnvironment

AEnvironment exposes an MCP server that enables tooling and environment capabilities for agentic RL and environment provisioning. It is built around the notion of Everything as Environment, allowing agents to call standardized tools (e.g., file operations, code execution, validations) and to interact with built-in environments such as TAU2 and Mini Terminal. To start the MCP server, use the AEnvironment CLI (aenv) which runs the MCP server within your project directory, exposing your tools and environment interfaces to clients that conform to the MCP protocol.

Once running, clients can discover available tools and environments, then invoke them through the MCP interface. The included examples illustrate how agents can read and write files, execute code, and perform validation tasks via registered tools. The server supports multi-environment composition, agent-as-environment scenarios, and integration with RL training workflows through its MCP-based tool and environment abstractions.

Prerequisites:

• Python 3.12+ installed on the host
• pip available

Step-by-step:

1. Create and activate a Python-based project directory for your MCP server (optional but recommended):

mkdir aenvironment-mcp
cd aenvironment-mcp

2. Install the AEnvironment package which provides the MCP server and tooling:

pip install aenvironment

3. (Optional) Initialize or configure your environment project if needed by your setup:

aenv init my-env  # or follow project-specific setup instructions

4. Start the MCP server locally (this runs the MCP interface so clients can connect):

aenv run

5. Verify the server is up and listening for MCP requests, then connect a client that uses the MCP protocol to enumerate tools and environments.

# Example connection (depending on client implementation)
python -c "import mcp_client; client = mcp_client.connect('http://localhost:port'); print(client.list_tools())"

Notes:

• The exact module and entry point may vary if your project wires the MCP server differently; the common pattern is to run the Python package/module that starts the MCP server (as shown with aenv run).
• Ensure your environment variables (if any) are set to match your deployment environment (e.g., ENVIRONMENT_NAME, NOTIFY_ON_START).

Tips and common issues:

• If the server fails to start, check Python version compatibility and that the aenvironment package is installed in the active environment.
• Ensure network ports used by the MCP server are not blocked by firewalls or in-use by other processes.
• When integrating with agents, leverage the built-in environments (TAU2, Mini Terminal, TerminalBench) as ready-made examples of tool invocation and environment handling.
• Use the Quick Start guide and the provided examples (mini-program, TAU2 RL) to understand how tools, functions, and rewards are registered and invoked.
• Environment variables can be used to customize startup behavior or integration with orchestration systems.