Meilisearch September 2026: MCP, search rules

Introduced a native Model Context Protocol (MCP) endpoint for direct querying by AI agents, enabling easier integration with AI coding assistants and RAG pipelines. Added experimental 'scale' action to search rules, allowing boosting, demoting, and hiding of documents based on filters or IDs without changing global ranking rules. Improved search performance in v1.51 by reducing disk reads (up to 5.4x faster on datasets with many fields) and in v1.49 by optimizing synonym loading (up to 13x faster on indexes with large synonym lists). Introduced a UI for setting up embedders, simplifying semantic and hybrid search configuration. Added self-serve disk performance add-ons with usage charts for Meilisearch Cloud. Enhanced foreign filters to work across shards in sharded setups (v1.53) with an increased retrieval limit (100 to 1,000 documents). Stabilized dumpless upgrades for self-hosted instances (v1.51). Added `performanceDetails` to multi-search requests to break down timing per query.