proxy-base-agent

The Proxy Base Agent (PBA) MCP server enables reliable, stateful agent execution powered by the Proxy State Engine (PSE). It exposes an interface to run the PBA as a Python-based agent that adheres to a defined state machine, guaranteeing correct tool usage and consistent state transitions during interactions with LLMs. When connected via MCP, you can dynamically introduce new tools and capabilities into the agent without restarting the server, taking advantage of PBA's dynamic updates through MCP integration. This makes it suitable for building robust automation workflows where tool calls must follow strict schemas and stateful plans are enforced at runtime.

To use the server, deploy the MCP server configuration and start the agent using the Quickstart flow described in the repository. The typical invocation after installation is to run the agent module with Python, which launches an interactive wizard to configure your LLM, tools, and workflow. Once running, the agent will guide you through selecting tools, defining states, and executing plans with guaranteed tool usage. You can also leverage MCP to connect additional external tools and services on-the-fly, with PSE enforcing correct transitions and argument schemas across all integrated tools.

Prerequisites:

• Python 3.8+ installed on your system
• Internet access to install dependencies

Step 1: Create a Python virtual environment (optional but recommended)

python -m venv venv
source venv/bin/activate  # On Windows use: venv\Scripts\activate

Step 2: Install the Proxy Base Agent package from PyPI

pip install proxy-base-agent

Step 3: Verify installation

python -m pip show proxy-base-agent

Step 4: Run the Quickstart to configure and start the agent

python -m agent

Step 5: If you intend to run via MCP, ensure your MCP server configuration exists (see mcp_config) and start your MCP runtime per your environment guidelines.

Tips and common considerations:

• The agent relies on the Proxy State Engine (PSE); ensure PSE-related dependencies are available in your environment when extending or customizing workflows.
• When connecting to external tools via MCP, you can reconfigure or add new tools on-the-fly; any state machine references will be updated accordingly by the MCP runtime.
• If you encounter tool-call argument mismatches, double-check the schema definitions within your state machines; PSE enforces schema compliance during generation and execution.
• The documentation and Quickstart provided by The Proxy Company offer deeper dives into core concepts, extending the agent, and API references for advanced use cases.
• This MCP server is Python-based; if you plan to containerize, you can adapt the run command to your container workflow and map in environment variables as needed.