Okta MCP Server

🔍 What is the Model Context Protocol?

<div align="left"> <p>The Model Context Protocol (MCP) is an open standard that enables AI models to interact with external tools and services in a structured, secure way. It provides a consistent interface for AI systems to discover and use capabilities exposed by servers, allowing AI assistants to extend their functionality beyond their training data.</p>

<p>Think of MCP as the "USB-C of AI integration" - just as USB-C provides a universal standard that allows various devices to connect and communicate regardless of manufacturer, MCP creates a standardized way for AI models to discover and interact with different services without custom integration for each one. This "plug-and-play" approach means developers can build tools once and have them work across multiple AI assistants, while users benefit from seamless integration without worrying about compatibility issues.</p>

<p><strong>Example:</strong> "Find all locked users in our Okta tenant, and create a spreadsheet in our IT Operations folder on Google Drive with their names, email addresses, and last login dates." <em>The AI uses Okta MCP Server to query locked users, then passes this data to Google Drive MCP Server to create the spreadsheet - all without custom coding.</em></p>

⚠️ IMPORTANT: Security & Limitations

Please read this section carefully before using Okta MCP Server.

🔄 Data Flow & Privacy

When you make a request, the interaction happens directly between the LLM and the Okta MCP tools - the client application is no longer in the middle. All data returned by these tools (including complete user profiles, group memberships, etc.) is sent to and stored in the LLM's context during the entire transaction for that conversation.

Key Privacy Considerations:

• The LLM (Claude, GPT, etc.) receives and processes all Okta data retrieved by the tools
• This data remains in the LLM's context for the duration of the conversation
• You must be comfortable with your Okta user data being processed by the LLM provider's systems
• Before using these tools, ensure you're comfortable with Okta data being sent to the AI model's servers

📊 Context Window Limitations

MCP is designed for lightweight workflows similar to Zapier, not bulk data operations.

Recommendation: Limit requests to fewer than 100 entities per transaction. Avoid operations that require fetching large datasets or multiple API calls.

Examples:

❌ Avoid these types of requests:

• "Fetch all 10,000 users from our Okta tenant and analyze their login patterns"
• "Find users who do not have Okta Verify enrolled as a factor"

✅ Better approaches:

• "Get the most recently created 20 users"
• "Find users who haven't logged in for 90+ days, limit to first 50 results"

💡 For larger data sets and complex queries: Consider using the Okta AI Agent for larger queries and data sets, The agent is being enhanced with similar "actionable" features to handle larger datasets and more complex scenarios in the very near future.

🚨 HTTP Transport Security Warning

The HTTP transport modes (both Streamable HTTP and SSE) have significant security risks:

• They open unauthenticated HTTP servers with full access to your Okta tenant
• No authentication or authorization is provided
• Anyone who can reach the network port can issue commands to your Okta environment
• EXTREMELY DANGEROUS when using remote HTTP access via mcp-remote

Best Practice: Only use the STDIO transport method (default mode) unless you have specific security controls in place and understand the risks.

🛠️ Available Tools

The Okta MCP Server currently provides the following tools:

User Management

• list_okta_users - Retrieve users with filtering, search, and pagination options
• get_okta_user - Get detailed information about a specific user by ID or login
• list_okta_user_groups - List all groups that a specific user belongs to
• list_okta_user_applications - List all application links (assigned applications) for a specific user
• list_okta_user_factors - List all authentication factors enrolled for a specific user

Group Operations

• list_okta_groups - Retrieve groups with filtering, search, and pagination options
• get_okta_group - Get detailed information about a specific group
• list_okta_group_members - List all members of a specific group
• list_okta_assigned_applications_for_group - List all applications assigned to a specific group

Application Management

• list_okta_applications - Retrieve applications with filtering, search, and pagination options
• list_okta_application_users - List all users assigned to a specific application
• list_okta_application_group_assignments - List all groups assigned to a specific application

Policy & Network Management

• list_okta_policy_rules - List all rules for a specific policy with detailed conditions and actions
• get_okta_policy_rule - Get detailed information about a specific policy rule
• list_okta_network_zones - List all network zones with IP ranges and configuration details

System Log Events

• get_okta_event_logs - Retrieve Okta system log events with time-based filtering and search options

Date & Time Utilities

• get_current_time - Get current UTC time in ISO 8601 format
• parse_relative_time - Convert natural language time expressions to ISO 8601 format

Additional tools for applications, factors, policies, and more advanced operations are on the roadmap and will be added in future releases.

🚀 Quick Start

Prerequisites

✅ Python 3.8+ installed on your machine ✅ Okta tenant with appropriate API access ✅ An MCP-compatible AI client (Claude Desktop, Microsoft Copilot Studio, etc.)

⚠️ Important Model Compatibility Note: Not all AI models work with this MCP server. Testing has only been performed with: - GPT-4.0 - Claude 3.7 Sonnet - Google-2.5-pro You must use latest model versions that explicitly support tool calling/function calling capabilities. Older models or models without tool calling support will not be able to interact with the Okta MCP Server.

🧠 Supported AI Providers

The Okta MCP Server supports multiple AI providers through its flexible configuration system. This allows you to connect to various large language models based on your specific needs and existing access.

Currently Supported Providers:

| Provider | Environment Variable | Description | |----------|---------------------|-------------| | OpenAI | AI_PROVIDER=openai | Connect to OpenAI API with models like GPT-4o. Requires an OpenAI API key. | | Azure OpenAI | AI_PROVIDER=azure_openai | Use Azure-hosted OpenAI models with enhanced security and compliance features. | | Anthropic | AI_PROVIDER=anthropic | Connect to Anthropic's Claude models (primarily tested with Claude 3.7 Sonnet). | | Google Vertex AI | AI_PROVIDER=vertex_ai | Use Google's Gemini models via Vertex AI. Requires Google Cloud service account. | | OpenAI Compatible | AI_PROVIDER=openai_compatible | Connect to any OpenAI API-compatible endpoint, such as Fireworks.ai, Ollama, or other providers that implement the OpenAI API specification. |

Installation

# Clone the repository
git clone https://github.com/fctr-id/okta-mcp-server.git
cd okta-mcp-server

# Create and activate a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows use: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

⚠️ NOTICE: If you clone this repository anew or pull updates, always make sure to re-run pip install -r requirements.txt to ensure all dependencies are up-to-date.

Configuration & Usage

Create a config file with your Okta settings:

To use the command line client (no memory), use the instructions below

# Copy the sample config
cp .env.sample .env

# Edit the env with your settings
# Required: Okta domain and API token and LLM settings

cd clients
python mcp-cli-stdio-client.py

To use MCP hosts like Claude Code, vsCode ...etc find the json config below

Supported Transports and Launching

The Okta MCP Server supports multiple transport protocols:

1. Standard I/O (STDIO) - Recommended

• Security: ✅ Direct communication through standard input/output streams
• Use case: Ideal for desktop AI assistants like Claude Desktop
• Performance: ✅ Lightweight and efficient
• Configuration: For Claude Desktop, add to claude_desktop_config.json:

  {
    "mcpServers": {
      "okta-mcp-server": {
        "command": "DIR/okta-mcp-server/venv/Scripts/python",
        "args": [
          "DIR/okta-mcp-server/main.py"
        ],
        "env": {
          "OKTA_CLIENT_ORGURL": "https://dev-1606.okta.com",
          "OKTA_API_TOKEN": "OKTA_API_TOKEN"
        }
      }
    }
  }

Replace DIR with your absolute directory path and OKTA_API_TOKEN with your actual token

2. Streamable HTTP Transport - Modern & Current Standard

Current Standard - Modern HTTP-based transport with advanced features:

• Features: ✅ Real-time event streaming, session management, resumability support
• Performance: ✅ Better scalability and connection handling
• Use case: Modern web applications and clients supporting HTTP streaming
• Security: ⚠️ Local HTTP server - secure in controlled environments

Starting the Streamable HTTP Server: ```bash

Start server with explicit risk acknowledgment

python main.py --http --iunderstandtherisks

Server will start on http://localhost:3000/mcp

Connect using streamable HTTP compatible clients

**Features:**
- ✅ **Real-time streaming** - Live progress updates during operations
- ✅ **Session management** - Maintains connection state
- ✅ **Event streaming** - Server-Sent Events for real-time notifications
- ✅ **Better error handling** - Detailed error responses
- ✅ **Modern protocol** - Based on latest MCP specifications

**For Streamable HTTP Client Testing:**

cd clients python mcp-cli-streamable-client.py ```

3. Remote HTTP Access - High Risk Advanced Use Only

⚠️ EXTREMELY DANGEROUS - READ CAREFULLY

For MCP clients that don't natively support remote connections, you can use mcp-remote via NPX:

Prerequisites:

• Node.js and NPM installed
• Okta MCP Server running in HTTP mode

Setup: ```bash

1. Install mcp-remote globally

npm install -g @anthropic/mcp-remote

2. Start your Okta MCP Server in HTTP mode

python main.py --http --iunderstandtherisks

3. Configure your MCP client (e.g., Claude Desktop)

**Claude Desktop Configuration:**

{ "mcpServers": { "okta-mcp-server": { "command": "npx", "args": [ "mcp-remote", "http://localhost:3000/mcp" ], "env": { "OKTA_CLIENT_ORGURL": "https://dev-1606.okta.com", "OKTA_API_TOKEN": "your_actual_api_token" } } } } ```

🚨 CRITICAL SECURITY WARNINGS:

• NEVER use in production environments
• NEVER expose the HTTP port (3000) to public networks
• ANYONE with network access can control your Okta tenant
• No authentication or authorization protection
• All Okta operations are exposed without restrictions
• Use only in isolated, secure development environments
• Consider this approach only if STDIO transport is absolutely not feasible

When might you need this approach:

• Testing MCP integrations that require HTTP transport
• Specific client applications that can't use STDIO
• Development scenarios requiring HTTP debugging
• NEVER for production or shared environments

4. Server-Sent Events (SSE) - Deprecated

⚠️ DEPRECATED: SSE transport is deprecated and not recommended for new implementations.

# Run in SSE mode (requires explicit risk acknowledgment)
python main.py --sse --iunderstandtherisks

• Use case: Legacy MCP clients that specifically require SSE (not recommended)
• Security: ⚠️ Same HTTP security risks as Streamable HTTP
• Recommendation: Use Streamable HTTP transport instead for all new implementations

5. Docker Deployment

The Okta MCP Server provides Docker images for all transport types, offering containerized deployment options.

Running Docker Containers

STDIO Transport (Recommended): For Claude Desktop or other MCP clients, configure to use the Docker container: ``json { "mcpServers": { "okta-mcp-server": { "command": "docker", "args": [ "run", "-i", "--rm", "-e", "OKTA_CLIENT_ORGURL", "-e", "OKTA_API_TOKEN", "fctrid/okta-mcp-server:stdio" ], "env": { "OKTA_CLIENT_ORGURL": "https://your-org.okta.com", "OKTA_API_TOKEN": "your_api_token" } } } } ``

Streamable HTTP Transport (Current Standard): ```bash

Start the HTTP container

docker run -d --name okta-mcp-http \ -p 3000:3000 \ -e OKTA_API_TOKEN=your_api_token \ -e OKTA_CLIENT_ORGURL=https://your-org.okta.com \ fctrid/okta-mcp-server:http

Configure your MCP client to connect to http://localhost:3000/mcp

**SSE Transport (Deprecated - Not Recommended):**

Start the SSE container (deprecated)

docker run -d --name okta-mcp-sse \ -p 3000:3000 \ -e OKTA_API_TOKEN=your_api_token \ -e OKTA_CLIENT_ORGURL=https://your-org.okta.com \ fctrid/okta-mcp-server:sse

Configure your MCP client to connect to http://localhost:3000/sse

**Building Images Locally:**

Build all variants

docker build --target stdio -t okta-mcp-server:stdio . docker build --target http -t okta-mcp-server:http . docker build --target sse -t okta-mcp-server:sse . ```

⚖️ Legal Stuff

Check out License.md for the fine print.

---

🌟 © 2025 Fctr Identity. All rights reserved. Made with ❤️ for the Okta and AI communities.