MCP Server Starter

/** * @module Tools/Search * @category Tools */ import { z } from "zod"; import { searchDocs } from "../shared/stub-data.js"; import type { RegisterableModule } from "../registry/types.js"; import type { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js"; /** * Implements the ChatGPT-required `search` tool. * * WHY: ChatGPT connectors and Deep Research expect a `search` tool that * returns a single content item of type `text` whose payload is a JSON string * with `{ results: Array<{ id, title, url }> }`. * * HOW: We query a tiny in-memory dataset for case-insensitive substring * matches and return the top results encoded as JSON in a single text item. * This mirrors the protocol shape without external dependencies. */ const searchModule: RegisterableModule = { type: "tool", name: "search", description: "Return JSON-encoded search results for stub documents", register(server: McpServer) { server.tool( "search", "Search stub documents and return minimal results (id, title, url)", { query: z.string().describe("Search query string"), }, async (args) => { const { query } = args; // Perform simple substring search against the stub dataset. const matches = searchDocs(query); // Map to the minimal result shape required by OpenAI docs. const results = matches.map((d) => ({ id: d.id, title: d.title, url: d.url })); return { content: [ { type: "text", // IMPORTANT: ChatGPT expects a JSON-encoded string, not an object text: JSON.stringify({ results }), }, ], }; } ); } }; export default searchModule;