Cloudflare Radar Beta Exposes Internet Data to MCP Agents

Cloudflare has launched the public beta of Radar Researcher, an autonomous conversational tool for analyzing global internet and security telemetry. Announced during Cloudflare Agents Week, the tool replaces manual dashboard filtering and API calls with natural language queries. Developers can now ask complex questions about global traffic, DNS routing, or connectivity drops and receive annotated charts, data summaries, and raw tables in return.

The system represents a shift from the simple query translators common in earlier iterations of Cloudflare’s AI Assistant. Radar Researcher is a fully autonomous research agent capable of executing multi-step data analysis against live datasets, including HTTP traffic, 1.1.1.1 resolver data, and Cloudflare Speed Test network quality metrics.

Technical Architecture

Radar Researcher serves as an internal demonstration of Cloudflare’s autonomous agent infrastructure, built entirely on the Cloudflare Developer Platform. The system relies on the Cloudflare Agents SDK running on Workers to manage logic and tool execution.

Reasoning is handled by Workers AI, utilizing a fallback chain of three different model families. The primary reasoning engine is the Kimi K2.7 model. The architecture is designed to automatically cascade inference requests to backup models if the primary family reaches capacity, maintaining uptime during intensive multi-step investigations.

Conversation state and memory are maintained using Stateful Storage. Each user thread is stored as a Durable Object backed by an isolated SQLite database. This architecture ensures persistent chat histories and allows streaming responses to continue executing server-side even if a user closes their browser window.

Agentic Features and Integration

The tool is designed to manage open-ended investigations. If asked about regional internet shutdowns, the agent determines the relevant time windows, queries multiple Radar endpoints, and synthesizes the data into interactive charts. Users can hover over timeline data points to view annotated events directly on the generated graphs.

Once a primary query is resolved, the agent suggests follow-up paths, such as comparing the initial region against neighboring countries or expanding the timeline to establish a historical baseline.

For developers building programmatic agents, Radar Researcher is accessible via a Model Context Protocol server. This allows third-party UIs, AI coding assistants, and custom agents to access Cloudflare Radar data using the same natural language interface without writing rigid API integration code.

The public beta is currently available for free on the Cloudflare Radar website. If you are building autonomous systems that require real-time internet telemetry, routing your data gathering through the new MCP server will eliminate the need to maintain static endpoint integrations.