docs-mcp-server

import { MimeTypeUtils } from "../../../utils/mimeTypeUtils"; import type { DocumentStore } from "../../DocumentStore"; import type { DbPageChunk } from "../../types"; import type { ContentAssemblyStrategy } from "../types"; const CHILD_LIMIT = 3; const PRECEDING_SIBLINGS_LIMIT = 1; const SUBSEQUENT_SIBLINGS_LIMIT = 2; /** * Assembly strategy that preserves the current behavior for markdown and text content. * * Uses broad context expansion (parents, siblings, children) and simple "\n\n" joining. * This strategy is optimized for prose content where broader context enhances understanding. */ export class MarkdownAssemblyStrategy implements ContentAssemblyStrategy { /** * Determines if this strategy can handle the given content type. * Handles markdown, HTML, plain text, and serves as fallback for unknown types. */ canHandle(mimeType?: string): boolean { // Handle undefined/unknown MIME types as fallback if (!mimeType) { return true; } // First, check if it's a structured type that should be handled by HierarchicalAssemblyStrategy if (MimeTypeUtils.isSourceCode(mimeType) || MimeTypeUtils.isJson(mimeType)) { return false; } // Handle markdown content if (MimeTypeUtils.isMarkdown(mimeType)) { return true; } // Handle HTML content if (MimeTypeUtils.isHtml(mimeType)) { return true; } // Handle plain text content if (MimeTypeUtils.isText(mimeType)) { return true; } // Accept as fallback for truly unknown types return true; } /** * Selects chunks using the current context expansion logic. * This replicates the existing behavior from DocumentRetrieverService.getRelatedChunkIds(). */ async selectChunks( library: string, version: string, initialChunks: DbPageChunk[], documentStore: DocumentStore, ): Promise<DbPageChunk[]> { const allChunkIds = new Set<string>(); // Process all initial chunks in parallel to gather related chunk IDs const relatedIdsPromises = initialChunks.map((doc) => this.getRelatedChunkIds(library, version, doc, documentStore), ); const relatedIdsResults = await Promise.all(relatedIdsPromises); // Add all related IDs to the set (automatically deduplicates) for (const relatedIds of relatedIdsResults) { for (const id of relatedIds) { allChunkIds.add(id); } } // Fetch all chunks and return them in sort_order const chunkIds = Array.from(allChunkIds); const chunks = await documentStore.findChunksByIds(library, version, chunkIds); return chunks; // Already sorted by sort_order in findChunksByIds } /** * Assembles chunks using simple "\n\n" joining (current behavior). */ assembleContent(chunks: DbPageChunk[]): string { return chunks.map((chunk) => chunk.content).join("\n\n"); } /** * Collects related chunk IDs for a single chunk using current context expansion logic. * This is a direct port of the logic from DocumentRetrieverService.getRelatedChunkIds(). */ private async getRelatedChunkIds( library: string, version: string, doc: DbPageChunk, documentStore: DocumentStore, ): Promise<Set<string>> { const id = doc.id; const relatedIds = new Set<string>(); // Add the original chunk relatedIds.add(id); // Parent const parent = await documentStore.findParentChunk(library, version, id); if (parent) { relatedIds.add(parent.id); } // Preceding Siblings const precedingSiblings = await documentStore.findPrecedingSiblingChunks( library, version, id, PRECEDING_SIBLINGS_LIMIT, ); for (const sib of precedingSiblings) { relatedIds.add(sib.id); } // Child Chunks const childChunks = await documentStore.findChildChunks( library, version, id, CHILD_LIMIT, ); for (const child of childChunks) { relatedIds.add(child.id); } // Subsequent Siblings const subsequentSiblings = await documentStore.findSubsequentSiblingChunks( library, version, id, SUBSEQUENT_SIBLINGS_LIMIT, ); for (const sib of subsequentSiblings) { relatedIds.add(sib.id); } return relatedIds; } }