oci-mcp-server

oci-mcp-server

A Python MCP (Multi-Client Protocol) server for Oracle Cloud Infrastructure (OCI) API Gateway management, with both programmatic and LLM-powered clients.

---

Features

• List API Gateways in a given OCI compartment.
• Get details for a specific API Gateway.
• MCP stdio transport for easy integration with LLM agents and other clients.
• LlamaIndex/Ollama client for natural language interaction with OCI API Gateway resources.

---

Quickstart

1. Prerequisites

• Python 3.11 (required)
• uv (fast Python package/dependency manager)
• OCI account with API Gateway access
• OCI session authentication files (~/.oci/sessions/DEFAULT/token and oci_api_key.pem)

2. Install uv

curl -Ls https://astral.sh/uv/install.sh | sh

Or, from PyPI:

# With pip.
pip install uv

# Or pipx.
pipx install uv

If installed via the standalone installer, uv can update itself to the latest version:

uv self-update

Or see uv installation docs for other methods.

3. Create and Activate a Virtual Environment

uv venv --python 3.11
source .venv/bin/activate

4. Install Dependencies

uv pip install -r pyproject.toml

Or, if you want to use uv directly:

uv pip install -e .

---

Cursor & MCP Integration

5. Install Cursor

Cursor is an AI-powered code editor that can natively interact with MCP servers.

To install Cursor:

• Download and install Cursor from the official website for your platform (macOS, Windows, Linux).
• Follow the installation instructions for your OS.

6. Update your mcp.json configuration

To connect Cursor to your custom MCP server, you need to update or create an mcp.json file in your project root. Use the provided template in src/mcp_template.json as a starting point.

Steps

1. Open Cursor Settings by navigating to Cursor > Settings... > Cursor Settings

2. Select MCP > + Add new global MCP server

You should also be able to edit mcp.json from vi ~/.cursor/mcp.json

Steps:

1. Copy the content of src/mcp_template.json to ~/.cursor/mcp.json (or to mcp.json from the editor):

2. Edit mcp.json and update the paths:

   • Replace <path to Python> with the absolute path to your Python 3.11 executable if using a virtual environment).
   • Replace <path to server.py> with the absolute path to your server.py script.

3. Save the file. Cursor will now be able to discover and connect to your MCP server.

---

Configuration

Edit src/config.py to set your OCI compartment and gateway OCIDs, region, and profile as needed:

PROFILE_NAME = 'DEFAULT'
REGION = 'ap-mumbai-1'
SERVICE_ENDPOINT = '<SERVICE ENDPOINT>'
COMPARTMENT_ID = '<COMPARTMENT_ID>'
GATEWAY_ID = '<GATEWAY_ID>' # For Programmatic Client

You must have valid OCI session files in ~/.oci/sessions/DEFAULT/.

You can do so by something like this oci session authenticate --tenancy-name <TENANCY_NAME> --profile-name DEFAULT --region $REGION

---

Running the MCP Server

From the project root: (I have found that its better to run the server in a separate terminal)

python src/server.py

This will start the MCP server using stdio transport, exposing the following tools:

• list_gateways_tool(compartment_id)
• get_gateway_tool(gateway_id)

---

Cursor Client (Agent Mode)

You do not need to run the MCP server for Cursor. As Cursor Agent will run it from the server resource provided in mcp.json

See src/mcp_template.json for an example MCP server config:

{
    "mcpServers": {
        "oci_api_gateway": { 
            "command": "<absolute path to Python (preferably from .venv)>",
            "args": [
                "<absolute path to server.py>"
            ]
        }
    }
}

---

In the Cursor editor, you can now interact with the MCP server using natural language queries.

For example, to list all the gateways in a specific compartment, you can type:

Example:

list all the gateways in ocid1.compartment.oc1..aaaaaaaai2zzsg.............................m4pjq

The Cursor agent will process your request and return the results from the MCP server.

Programmatic Client

The Programmatic Client allows you to test and interact with your custom MCP tools directly using Python code, without involving an AI agent. It's useful for verifying tool functionality and debugging server responses in a straightforward, scriptable way.

See src/client_stdio.py for a simple example:

python src/client_stdio.py

This will:

• Connect to the MCP server
• List available tools
• Call the list_gateways_tool and get_gateway_tool methods and print results

---

WIP: LlamaIndex + Ollama Client

The LlamaIndex + Ollama Client is a work in progress (WIP) that aims to enable natural language interaction with your MCP tools using an self hosted LLM agent

See src/client_llama_react.py for an advanced example using LlamaIndex and Ollama (Llama 3.2):

• Make sure you have Ollama running locally with the llama3.2 model.
• Edit the server_params path in client_llama_react.py if needed.

python src/client_llama_react.py

This will:

• Connect to the MCP server
• Wrap the tools for LlamaIndex
• Run a natural language query using the LLM

---

Development

• All source code is in the src/ directory.
• Dependencies are managed via pyproject.toml.
• To add new tools, edit src/server.py and implement logic in src/gateway_services.py(for gateway resources).

---

Troubleshooting

• Ensure your OCI session files are present and valid.
• Use Python 3.11 and the provided virtual environment.
• For LlamaIndex/Ollama, ensure the Ollama server is running and accessible.