Introducing the Oso MCP Server: Your new Authorization Co-pilot

Today, we are launching the official Oso MCP server. You can use this server to supply your favorite AI tools (such as Cursor or Claude) with access to the Oso Cloud API. 

The Oso MCP is a useful companion when learning Polar, debugging unexpected behavior in your application, and accelerating the development of new features. It provides a complete interface for managing Oso Cloud authorization policies and data.
‍

What is an MCP Server?

MCP (Model Context Protocol) standardizes how LLMs interact with external tools by creating a common protocol that both AI clients (like Cursor) and tool providers can implement, eliminating the need for developers to duplicate tool implementations across projects. Instead of each app hand-picking specific tools, MCP enables a plugin-style ecosystem where users can bring their own tools to any MCP-compatible client. Beyond tools, MCP also supports resources (arbitrary data like files or database records) and prompts (reusable templates) to provide richer context for LLMs.

The Oso Cloud MCP server connects your AI tools to the Oso platform, enabling them to explain your policies, answer specific questions about facts, or write tests on your behalf.
‍

Setup

First, create an API key in your project dashboard to authenticate your agent’s requests. 

Then, you can setup the Oso MCP on most AI clients using the following JSON:

{
 "mcpServers": {
   "oso": {
     "command": "oso-cloud",
     "args": ["experimental", "mcp"],
     "env": {
       "OSO_AUTH": "<your_api_key>"
     }
   }
 }
}

Oso MCP Server Demo Video

Get Started with Oso Cloud

We've built the Oso MCP server to bridge the gap between AI tools and your authorization systems, letting you focus on building secure applications instead of context-switching between policy debugging and development tools.

The Model Context Protocol is evolving with proposals like the new Streamable HTTP transport that supports fully stateless servers without requiring long-lived connections. We're following these developments closely and evaluating how they might benefit the Oso MCP experience.

If you run into issues or have ideas for new tools we should add, leave us some feedback. We're particularly interested in hearing about your experiences with policy testing, fact querying, and authorization debugging as we continue to refine the server's capabilities.

Check out our docs for the latest updates and examples of what you can build with Oso MCP.