Kai MCP Server: Use Kai From Any AI Assistant

Kai now exposes a Model Context Protocol (MCP) server, giving AI assistants and development tools direct access to Kai's full capabilities. Connect your IDE, agent framework, or custom tooling to scan code, manage vulnerabilities, run evolutions, and more through a standardized protocol.

What Is MCP?

The Model Context Protocol is an open standard that lets AI assistants interact with external tools and services. Instead of copy-pasting results between Kai and your AI assistant, MCP lets them talk directly. Your assistant can trigger scans, read findings, create issues, and start evolutions without you ever leaving the conversation.

Getting Started

Add Kai to your MCP client with a single configuration entry. For Claude Desktop:

{
  "mcpServers": {
    "kai": {
      "url": "https://production.kai-backend.dria.co/mcp"
    }
  }
}

For Cursor, VS Code Copilot, or any other MCP-compatible client, add the same server URL in your MCP settings. On first connection, you'll be prompted to authorize with GitHub through a standard OAuth 2.1 flow. After that, sessions are managed automatically.

29 Tools Across 8 Categories

The MCP server exposes Kai's full API surface as structured tools that any AI assistant can invoke.

Workspaces and Repositories let you list workspaces, add GitHub repositories, browse file trees, and read file contents directly through your assistant.

Security Scans give you full control over Kai's scanning pipeline. Check available scan tiers and their agent configurations, start a scan, monitor progress with real-time status updates and agent iteration logs, or abort a running scan.

Exploits surface vulnerability findings from completed scans. List all findings with their severity and status (candidate, verified, verified and fixed), or drill into a specific exploit for the full hypothesis, verification details, and suggested fixes.

Evolutions connect to Kai Evolve for autonomous code optimization. Start evolution tasks, track iteration history and progress, review generated programs, or abort a running evolution.

Evaluators let you manage the evaluation criteria for evolutions. List existing evaluators, view their details, or generate new evaluators for specific code scopes using LLM-based analysis.

Integrations bridge findings to your issue tracker. Create GitHub issues or Jira tickets directly from exploit findings, without leaving your AI assistant.

Billing provides visibility into credit balances and transaction history for your workspace.

Example: Scan and Triage in Claude Desktop

Here's a typical workflow entirely within an AI conversation:

1. Ask your assistant to list your Kai workspaces and pick the right one
2. Find the repository you want to scan
3. Check available scan tiers and start a scan
4. Monitor progress as the scan runs
5. When complete, review the findings sorted by severity
6. For critical findings, ask the assistant to create GitHub issues directly

The entire loop happens in a single conversation. No tab switching, no copy-pasting, no context loss.

Example: Start an Evolution From Your IDE

If you're working in Cursor or VS Code with the MCP server connected:

1. Ask your assistant to start an evolution on a specific repository
2. Monitor the evolution's progress and iteration history
3. Review the generated programs and their fitness scores
4. Apply the best solution directly in your editor

Supported Clients

Any client that implements the MCP specification can connect to Kai's server. We've verified support with:

• Claude Desktop (Claude Code, Claude for Mac)
• Cursor
• VS Code (GitHub Copilot with MCP support)
• Custom agents built on the MCP SDK

The server uses SSE (Server-Sent Events) transport and follows the MCP specification for tool discovery and invocation.

Learn More

Visit our MCP server documentation for the full tool reference, or connect and start exploring. Your AI assistant will discover all available tools automatically.