If you're building Plesk extensions and tired of manually searching across five different documentation sources (admin guide, REST API, CLI, PHP SDK, JS SDK), this server consolidates them into a single semantic search endpoint. It uses hybrid retrieval with vector search via LanceDB, full-text search with Tantivy, and cross-encoder reranking to surface relevant chunks in under a second. The search_plesk_unified tool handles the heavy lifting, while get_file_content and resolve_references let you drill into specific files or trace symbol usage. It runs entirely offline after initial indexing, uses AST-aware chunking for code boundaries, and can optionally synthesize answers with inline citations. Three model profiles (light, medium, full-tq) let you trade accuracy for latency.
claude mcp add --transport stdio barateza-mcp-plesk-dev-docs -- uvx mcp-plesk-dev-docs