Limitless MCP Server

Instructions

Implementation Reference

• src/server.ts:758-794 (registration)

  Tool registration for 'limitless_search_conversations_about', including thin wrapper handler that delegates to the main implementation in AdvancedSearch.searchConversationsAbout
  // Advanced Search Tool server.tool("limitless_search_conversations_about", "Advanced search across ALL lifelogs (not just recent) with intelligent context, relevance scoring, and comprehensive filtering options. Perfect for finding historical discussions.", AdvancedSearchArgsSchema, async (args, _extra) => { try { let timeRange = undefined; if (args.time_expression) { const parser = new NaturalTimeParser({ timezone: args.timezone }); timeRange = parser.parseTimeExpression(args.time_expression); } const searchOptions: SearchOptions = { includeContext: args.include_context, maxResults: args.max_results, searchInSpeaker: args.speaker_name, searchInContentType: args.content_types, timeRange }; const results = await AdvancedSearch.searchConversationsAbout( limitlessApiKey, args.search_term, searchOptions ); const resultText = results.length === 0 ? `No conversations found about "${args.search_term}".` : `Found ${results.length} relevant conversation(s) about "${args.search_term}":\n\n${JSON.stringify(results, null, 2)}`; return { content: [{ type: "text", text: resultText }] }; } catch (error) { const errorMessage = error instanceof Error ? error.message : String(error); return { content: [{ type: "text", text: `Error searching conversations: ${errorMessage}` }], isError: true }; } } );
• src/search-and-analytics.ts:33-110 (handler)

  Core handler function implementing the search logic: fetches lifelogs across time range, identifies relevant content nodes containing the search term (with speaker/content type filters), computes relevance scores, extracts surrounding context, generates summaries, and returns top scored results.
  static async searchConversationsAbout( apiKey: string, searchTerm: string, options: SearchOptions = {} ): Promise<TopicSearchResult[]> { const { includeContext = true, contextLines = 2, maxResults = 20, minRelevanceScore = 0.3, searchInSpeaker, searchInContentType, timeRange } = options; // Fetch all relevant lifelogs const lifelogParams: any = {}; if (timeRange) { lifelogParams.start = timeRange.start; lifelogParams.end = timeRange.end; lifelogParams.timezone = timeRange.timezone; } // Use a large limit to search through more history lifelogParams.limit = 1000; lifelogParams.includeMarkdown = true; lifelogParams.includeHeadings = true; const allLifelogs = await getLifelogs(apiKey, lifelogParams); const results: TopicSearchResult[] = []; const searchTermLower = searchTerm.toLowerCase(); for (const lifelog of allLifelogs) { const matchingNodes = this.findRelevantNodes( lifelog, searchTermLower, searchInSpeaker, searchInContentType ); if (matchingNodes.length === 0) continue; // Calculate relevance score const relevanceScore = this.calculateRelevanceScore(matchingNodes, searchTermLower); if (relevanceScore < minRelevanceScore) continue; // Get context if requested let contextBefore: LifelogContentNode[] = []; let contextAfter: LifelogContentNode[] = []; if (includeContext && lifelog.contents) { const { before, after } = this.extractContext( lifelog.contents, matchingNodes, contextLines ); contextBefore = before; contextAfter = after; } results.push({ lifelog, relevantNodes: matchingNodes, contextBefore, contextAfter, relevanceScore, summary: this.generateSearchSummary(lifelog, matchingNodes, searchTerm) }); } // Sort by relevance and return top results return results .sort((a, b) => b.relevanceScore - a.relevanceScore) .slice(0, maxResults); }
• src/server.ts:237-245 (schema)

  Zod schema defining the input arguments for the tool registration.
  const AdvancedSearchArgsSchema = { search_term: z.string().describe("Text to search for across ALL lifelogs (not just recent ones)."), time_expression: z.string().optional().describe("Natural time range like 'this week', 'past month', 'today' to limit search scope."), speaker_name: z.string().optional().describe("Filter results to specific speaker/participant."), content_types: z.array(z.string()).optional().describe("Filter by content node types like ['heading1', 'heading2', 'blockquote']."), include_context: z.boolean().optional().default(true).describe("Include surrounding context for better understanding."), max_results: z.number().optional().default(20).describe("Maximum number of results to return."), timezone: z.string().optional().describe("IANA timezone for time calculations."), };
• src/search-and-analytics.ts:18-26 (schema)

  TypeScript interface defining options passed to the search handler, mapping from tool args.
  export interface SearchOptions { includeContext?: boolean; contextLines?: number; maxResults?: number; minRelevanceScore?: number; searchInSpeaker?: string; searchInContentType?: string[]; timeRange?: TimeRange; }
• src/search-and-analytics.ts:112-136 (helper)

  Helper method to find content nodes in a lifelog matching the search term and filters.
  private static findRelevantNodes( lifelog: Lifelog, searchTermLower: string, searchInSpeaker?: string, searchInContentType?: string[] ): LifelogContentNode[] { if (!lifelog.contents) return []; return lifelog.contents.filter(node => { // Content match const hasContentMatch = node.content?.toLowerCase().includes(searchTermLower); // Speaker filter if (searchInSpeaker && node.speakerName !== searchInSpeaker) { return false; } // Content type filter if (searchInContentType && !searchInContentType.includes(node.type)) { return false; } return hasContentMatch; }); }