cuga-agent

CUGA is a state-of-the-art generalist agent designed for enterprise-style tasks. The MCP (Model Context Protocol) integration allows you to run CUGA as a programmable server that exposes its capabilities to tooling and workflows via standardized interfaces. The Python-based CUGA agent can be started as a server (for example, via the cuga_agent package) and then leveraged within your orchestration environment to perform autonomous decision-making, tool usage, and policy-driven governance using the enterprise SDK. CUGA supports hybrid task execution, human-in-the-loop approvals, and pluggable tool integration through OpenAPI specs, Langflow, and the MCP ecosystem. When running as an MCP server, you can connect it to your tooling layer to orchestrate planning, execution, and monitoring in a scalable, enterprise-friendly manner. The server can be configured to use production-safe policies (Intent Guard, Playbook, Tool Approval, Tool Guide, Output Formatter) to govern actions, respect safety gates, and provide audit trails for decision points.

To use the server, install the CUGA agent package and run the module that exposes the MCP endpoint. The agent can connect to REST APIs, local Python functions, and OpenAPI-described tools. Through the SDK and the policies, you can tailor behavior to domain needs, configure safety gates, and manage workflows. Typical usage involves starting the MCP-enabled CUGA instance, registering tools (APIs, web actions, or internal functions), and invoking tasks that the agent will plan, execute, and report back on with structured outputs.

Prerequisites:

• Python 3.12+
• Access to a Python package index (PyPI) or a prebuilt wheel for the CUGA agent

Installation steps:

1. Create and activate a Python virtual environment (optional but recommended):

   python3 -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate

2. Install the CUGA agent package from PyPI or your internal index:

   pip install cuga-agent

3. Verify installation by listing the package or running a quick help:

   python -m cuga_agent --help

4. Prepare a configuration for MCP (see mcp_config below) and ensure any required environment variables are set.

5. Start the MCP-enabled CUGA agent server:

   python -m cuga_agent

Notes:

• If your deployment uses a specific entrypoint or custom launcher, adjust the module name accordingly (e.g., -m cuga_agent.server).
• If you’re behind a firewall or require proxy settings, configure environment variables (HTTP_PROXY, HTTPS_PROXY) as needed.

Tips and common issues:

• Ensure Python 3.12+ is installed and that your environment is using the correct Python interpreter (especially in virtual environments).
• If the agent relies on external tools or Playwright for hybrid tasks, install browser support as described in the setup guide (e.g., playwright install chromium).
• For policy-driven governance, configure the enterprise SDK or UI as documented in the SDK/Policies guides to enable Intent Guard, Playbook, Tool Approval, Tool Guide, and Output Formatter controls.
• When integrating via MCP, clearly define the tools and endpoints your agent will access, and ensure API keys or credentials are securely managed (use env vars or secret management).
• If you encounter import errors, verify that the cuga_agent package exposes the expected module and that your PYTHONPATH includes the installed site-packages directory.