Cloudflare AutoRAG MCP Server

# Changelog All notable changes to the Cloudflare AutoRAG MCP Server will be documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). ## [2.0.0] - 2025-01-25 ### Added - **Multi-AutoRAG support** for managing multiple AutoRAG instances - New `list_autorags` tool to list all configured AutoRAG instances - New `get_current_autorag` tool to get the current default AutoRAG instance - Added `autorag_name` parameter to all search tools (`autorag_basic_search`, `autorag_rewrite_search`, `autorag_ai_search`) - Support for configuring multiple instances via environment variables in wrangler.toml - **Enhanced schema documentation** per MCP best practices - Added query length limits (max 10,000 characters) to all search tools - Added explicit score range documentation (0.0-1.0) in tool descriptions - Added parameter constraints for `max_num_results` (1-50) ### Changed - Version bumped to 2.0.0 (major version due to new feature set) - All search tools now accept optional `autorag_name` parameter to specify which AutoRAG instance to use - Enhanced test suite to cover multi-AutoRAG functionality - Improved error handling for invalid AutoRAG instance names ### Removed - VECTORIZE binding from wrangler.toml (not needed - AutoRAG handles vector access internally) ### Configuration - Configure multiple AutoRAG instances in wrangler.toml: ```toml [vars] AUTORAG_NAME = "default-autorag" AUTORAG_INSTANCES = "default-autorag,secondary-autorag,specialized-autorag" AUTORAG_DESCRIPTIONS = "Main knowledge base,Secondary knowledge base,Specialized documents" ``` ### Technical Notes - Model selection features were investigated but not implemented (requires REST API, not available via Workers bindings) - Multi-AutoRAG selection works entirely through Workers bindings - Default AutoRAG instance specified by `AUTORAG_NAME` environment variable - Additional instances configured as comma-separated lists - Maintains full backward compatibility - `autorag_name` parameter is optional - VECTORIZE binding is not required - AutoRAG manages vector index access internally through the AI binding ## [1.2.0] - 2025-01-24 ### Added - **Cursor-based pagination support** for AI search tool only - New `cursor` parameter accepts pagination tokens from previous responses - AI search responses include `nextCursor` field for MCP compliance - Maintains backward compatibility (cursor is optional) ### Changed - Version bumped to 1.2.0 to reflect new feature addition - AI search responses now include `nextCursor` when more results are available ### Technical Notes - Implements MCP-compliant cursor-based pagination pattern - Only `autorag_ai_search` supports pagination (uses Cloudflare's `aiSearch` binding method) - Basic and rewrite search use `search` binding method which doesn't support pagination - Uses opaque cursor tokens for stateless pagination - Enables iterative retrieval for large result sets - Helps prevent LLM token window overflows ## [1.1.3] - 2025-01-24 ### Changed - **BREAKING**: Removed `filters` parameter from all search tools - Improved error messages when filter parameters are attempted ### Added - Custom error message directing users to REST API for filter functionality - Comprehensive test suite with local artifact storage ### Fixed - Eliminated confusing "Invalid input" errors when using filters - Removed non-functional filter documentation ### Technical Notes - Filters are not supported in Cloudflare Workers bindings, only in REST API - This change aligns the API surface with actual Workers binding capabilities ## [1.1.2] - 2025-01-24 ### Changed - Attempted to fix filter format to use comparison operators ### Discovered - Workers bindings don't support filters at all (REST API only) ## [1.1.1] - 2025-01-22 ### Added - `include_ai_response` parameter to AI search tool - Default score threshold of 0.5 for all search tools - Comprehensive parameter validation ## [1.1.0] - 2025-01-20 ### Added - Three distinct search tools: - `autorag_basic_search` - Pure vector search - `autorag_rewrite_search` - Search with query rewriting - `autorag_ai_search` - Full AI-powered search - Boolean parameter support ## [1.0.0] - 2025-01-15 ### Added - Initial release - Basic AutoRAG search functionality - MCP server implementation - Cloudflare Workers deployment support