@cloudflare/mcp-server-cloudflare

Implements the Model Context Protocol (MCP) specification as a production-grade server deployed on Cloudflare Workers, using HTTP streaming via /mcp endpoint with streamble-http transport for bidirectional communication between LLMs and Cloudflare services. Handles tool discovery, prompt templates, and resource management through standardized MCP message framing with automatic serialization/deserialization of tool schemas and responses.

Provides two authentication pathways: OAuth 2.0 flow for user-based access (interactive authorization with Cloudflare account) and API token mode for programmatic access (service-to-service authentication). Implements secure credential validation, token refresh, and user state management through Durable Objects for session persistence, with automatic credential injection into downstream Cloudflare API calls.

Implements a specialized MCP server for searching Cloudflare documentation and code examples using semantic search powered by Vectorize embeddings. Enables LLMs to find relevant documentation sections, API examples, and best practices based on natural language queries, with support for filtering by documentation category (Workers, Pages, API, etc.) and code language.

Exposes Cloudflare Browser Rendering capabilities through MCP tools for rendering web pages, capturing screenshots, and extracting page content. Implements headless browser automation with support for JavaScript execution, form interaction, and dynamic content rendering, providing LLMs with the ability to analyze visual content and interact with web applications.

Provides shared packages (@repo/mcp-common, @repo/mcp-observability, @repo/eval-tools) that all MCP servers depend on for authentication, metrics collection, and testing. Implements centralized observability through structured logging, distributed tracing, and metrics aggregation, with support for monitoring tool execution latency, error rates, and authentication failures across all servers.

Implements a production monorepo structure using pnpm workspaces for dependency management and Turbo for build orchestration, enabling efficient development and deployment of 15+ independent MCP servers. Provides shared build configuration, testing infrastructure (Vitest), and deployment pipelines that reduce duplication and ensure consistency across all servers.

Maintains a centralized registry of 100+ tools across 15+ specialized MCP servers (Workers Observability, DNS Analytics, AI Gateway, etc.), each with JSON Schema definitions for parameters and return types. Implements automatic tool discovery, schema validation, and routing to the appropriate server based on tool namespace, with support for tool categorization (Common Tools, Container Management, Observability, Workers Management, AI & Data Tools).

Exposes Cloudflare Workers runtime observability through MCP tools that query Analytics Engine, tail real-time logs, retrieve error traces, and analyze performance metrics. Implements direct integration with Cloudflare's Analytics Engine for structured query execution and Durable Objects for log streaming, providing LLMs with visibility into Worker execution, CPU time, memory usage, and request/error patterns.

Provides MCP tools for querying DNS analytics data through Cloudflare's Analytics Engine, enabling LLMs to analyze query patterns, identify top domains, detect DNS anomalies, and retrieve DNSSEC validation metrics. Implements time-series aggregation queries with configurable granularity (1m, 5m, 1h, 1d) and supports filtering by query type, response code, and geographic origin.

Exposes Cloudflare AI Gateway capabilities through MCP tools, enabling LLMs to configure model routing policies, analyze gateway metrics, and manage rate limiting rules. Implements schema-based configuration for routing strategies (load balancing, failover, A/B testing) and provides visibility into gateway performance, cache hit rates, and model availability across multiple LLM providers.

Exposes Cloudflare Workers deployment pipeline through MCP tools for triggering builds, monitoring deployment status, retrieving build logs, and managing deployment rollbacks. Implements integration with Cloudflare's build system to track compilation status, asset bundling, and deployment propagation across the global edge network, with support for preview deployments and production rollouts.

Implements automatic Retrieval-Augmented Generation (RAG) capabilities through MCP tools that index documents into Cloudflare Vectorize, perform semantic search, and retrieve relevant context for LLM prompts. Uses vector embeddings to enable semantic similarity matching beyond keyword search, with support for document chunking, metadata filtering, and relevance scoring.

Provides MCP tools for executing code in isolated sandbox containers and analyzing code structure through DEX (Dependency Explorer). Implements container lifecycle management (create, execute, cleanup) with resource limits (CPU, memory, timeout), and provides code analysis capabilities including dependency extraction, AST parsing, and security scanning.

Exposes Cloudflare Logpush capabilities through MCP tools for configuring log delivery, querying audit logs, and analyzing security events. Implements integration with Cloudflare's audit log system to retrieve user actions, API calls, and security events with filtering by timestamp, action type, and resource, supporting both real-time streaming and historical log queries.