MCP TypeScript Demo Server

import search from 'arxiv-api-ts'; import { XMLParser } from 'fast-xml-parser'; interface PaperInfo { title: string; authors: string[]; summary: string; pdf_url: string; published: string; } interface PapersInfo { [key: string]: PaperInfo; } export async function search_papers(topic: string, max_results: number = 5): Promise<PapersInfo> { /** * Search for papers on arXiv based on a topic and return their information. * * @param topic - The topic to search for * @param max_results - Maximum number of results to retrieve (default: 5) * @returns Promise resolving to object with paper IDs as keys and paper info as values */ try { // Use arxiv-api-ts to find the papers const response = await search.default({ searchQueryParams: [{ include: [{ name: topic }] }], maxResults: max_results, sortBy: 'relevance' }); const papersInfo: PapersInfo = {}; const papers = response.entries || []; for (const paper of papers) { const paperId = paper.id?.split('/').pop() || paper.id; let authors: string[] = []; let authorsArray: any[] = []; if (paper.authors) { if (Array.isArray(paper.authors)) { authorsArray = paper.authors; } else { authorsArray = [paper.authors]; } } authors = authorsArray.map((author: any) => { return author && author.name ? author.name : ''; }).filter(name => name !== ''); const paperInfo: PaperInfo = { title: paper.title || '', authors: authors, summary: paper.summary || '', pdf_url: paper.pdf_url || '', published: paper.published ? new Date(paper.published).toISOString().split('T')[0] : '' }; papersInfo[paperId] = paperInfo; } return papersInfo; } catch (error) { console.error('Error searching papers:', error); throw error; } } export async function extract_info(paper_id: string): Promise<string> { /** * Search for information about a specific paper by ID from arXiv. * * @param paper_id - The ID of the paper to look for * @returns Promise resolving to JSON string with paper information if found, error message if not found */ try { // Fetch from arXiv API using id_list const apiUrl = `http://export.arxiv.org/api/query?id_list=${encodeURIComponent(paper_id)}`; const response = await fetch(apiUrl); const xml = await response.text(); const parser = new XMLParser({ ignoreAttributes: false, attributeNamePrefix: '', trimValues: true }); const parsed = parser.parse(xml); const entries = parsed.feed && parsed.feed.entry ? (Array.isArray(parsed.feed.entry) ? parsed.feed.entry : [parsed.feed.entry]) : []; if (entries.length === 0) { return `No information found for paper ID ${paper_id}.`; } const entry = entries[0]; // Return all info from the API for this entry return JSON.stringify(entry, null, 2); } catch (error) { console.error('Error extracting paper info:', error); return `Error: ${error}`; } } // --- Prompt Templates --- /** * Generates a detailed prompt for tool selection and argument extraction. * @param toolList - A string describing the available tools and their arguments. * @param userQuery - The user's query string. * @returns The full prompt string for the LLM. */ export function getToolSelectionPrompt(toolList: string, userQuery: string): string { return `You are an expert assistant that helps users interact with a set of tools. Your job is to select the most appropriate tool for a given user query and extract the required arguments for that tool. You must always use the exact argument names and types as specified in the tool definitions.\n\nAvailable tools:\n${toolList}\n\nEach tool is described with its name, a description, and the arguments it accepts. For each argument, the type and a description are provided.\n\nGiven the following user query, select the most appropriate tool and extract the arguments from the query. Respond ONLY with a JSON object in the following format:\n\n{ \"tool\": <tool_name>, \"arguments\": { <argument_name>: <value>, ... } }\n\n- Use the exact argument names and types as specified.\n- If an argument is optional and not provided by the user, omit it from the arguments object.\n- Do not include any explanation or extra text.\n\nUser query: \"${userQuery}\"\n\nExamples:\n\nExample 1:\n- Tools:\n - search_papers: Search for papers on arXiv. Arguments: { name (string): Topic to search for, max_results (number, optional): Maximum number of results }\n - extract_info: Extract information from a paper. Arguments: { paper_id (string): ID of the paper }\n- User query: \"Find me the latest papers about quantum computing.\"\n- Output: { \"tool\": \"search_papers\", \"arguments\": { \"name\": \"quantum computing\" } }\n\nExample 2:\n- User query: \"Show me details for paper 1234.5678\"\n- Output: { \"tool\": \"extract_info\", \"arguments\": { \"paper_id\": \"1234.5678\" } }\n\nExample 3:\n- User query: \"List 10 papers about deep learning\"\n- Output: { \"tool\": \"search_papers\", \"arguments\": { \"name\": \"deep learning\", \"max_results\": 10 } }\n\nIf the user query does not match any tool, respond with:\n{ \"tool\": null, \"arguments\": {} }\n`; }