The ACI Unified MCP server provides a single entry point to access and orchestrate hundreds of tools through a consistent model-context protocol. It exposes your configured tool-suite as an MCP server, enabling AI agents to discover, authorize, and call tools via a unified interface. With ACI.dev’s MCP server, you can leverage multi-tenant authentication, dynamic tool discovery, and granular permission checks, while keeping tool integrations centralized and auditable. Once running, you can connect your agent framework or your own orchestrator to the MCP endpoint and query available tools, invoke functions, and receive structured results, all while benefiting from ACI.dev’s authentication flows and logging. In practice, you’ll configure tool access (OAuth scopes, secrets, and permissions) once, then rely on the MCP server to route calls to the appropriate tool handlers and return standardized responses suitable for LLMs or agents.

Usage typically involves starting the MCP server, pointing your client to the MCP endpoint, and then issuing tool discovery and call requests. You can combine it with ACI.dev’s python SDK or Typescript SDK to generate clients, or directly issue MCP calls from your agent’s planning layer. The server supports multi-tenant setups and can be deployed behind common hosting platforms to power your VibeOps workflows, automated provisioning tasks, and general tool orchestration across services.

Prerequisites:

• Python 3.8+ (recommended 3.9+)
• pip (Python package manager)
• Access to install Python packages from PyPI

Step-by-step installation:

1. Create a virtual environment (recommended): python -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate

2. Install the MCP server package (adjust package name if needed): pip install aci-mcp

3. Verify installation: python -m aci.mcp.server --help

4. Prepare a basic configuration file (optional but recommended) and start the server:

   Example config file: mcp_config.json

   { "mcpServers": { "aci": { "command": "python", "args": ["-m", "aci.mcp.server"] } } }

   Start server (if using a config-driven start):

   python -m aci.mcp.server --config mcp_config.json

5. Run with environment variables as needed for secrets and host binding (see additional_notes): export ACI_MCP_PORT=8080 export ACI_MCP_LOG_LEVEL=INFO python -m aci.mcp.server --config mcp_config.json

Notes and tips:

• If you’re deploying behind a reverse proxy or load balancer, ensure proper health checks and TLS termination are configured.
• Environment variables for secrets and OAuth credentials should be managed securely (e.g., via a secret manager or CI/CD vault).
• Common issues include module not found errors after installation; verify your virtual environment is active and that the package name matches the MCP server you intend to run.
• You can customize discovery and permissions via the MCP server configuration (e.g., per-tool scopes, tenant policies).
• If you need Python SDK compatibility, you can pair the MCP server with ACI.dev’s Python SDK to simplify client calls from your agents.
• For local testing, run with DEBUG or set ACI_MCP_LOG_LEVEL=DEBUG to get more verbose logs to diagnose issues.