import { GoogleGenAI, mcpToTool } from "@google/genai";
import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";
import type { CallToolRequest } from "@modelcontextprotocol/sdk/types.js";
import { afterEach, beforeEach, describe, expect, test } from "vitest";
import type { SearchQueuesResponse } from "../../src/tools/searchQueues.js";
import {
type CallToolInterception,
type CallToolResult,
callToolInterceptor,
prettyPrintResults,
spyOnClientCallTool,
toolUsageEvaluator,
} from "./utils/index.js";
const timeout = 30_000;
const model = "gemini-2.5-flash";
const searchQueueInterceptor: CallToolInterception = {
shouldIntercept: (call: CallToolRequest["params"]) => {
return call.name === "search_queues";
},
intercept: (): Promise<CallToolResult> => {
const response: SearchQueuesResponse = {
queues: [
{
name: "Lucas_Test_Queue",
id: "c1c893ed-b765-4361-abc3-7bb3f5b7d6e5",
memberCount: 0,
},
],
pagination: {
pageNumber: 1,
pageSize: 100,
totalPages: 1,
totalMatchingQueues: 1,
},
};
return Promise.resolve({
content: [
{
type: "text",
text: JSON.stringify(response),
},
],
});
},
};
const interceptors = [searchQueueInterceptor];
describe(
"Evaluate efficient tool use",
() => {
let client: Client;
let ai: GoogleGenAI;
beforeEach(async () => {
const transport = new StdioClientTransport({
command: "npm",
args: ["run", "start"],
env: {
GENESYSCLOUD_REGION: process.env.GENESYSCLOUD_REGION ?? "",
GENESYSCLOUD_OAUTHCLIENT_ID:
process.env.GENESYSCLOUD_OAUTHCLIENT_ID ?? "",
GENESYSCLOUD_OAUTHCLIENT_SECRET:
process.env.GENESYSCLOUD_OAUTHCLIENT_SECRET ?? "",
},
});
client = new Client({
name: "genesys-cloud-mcp-client",
version: "1.0.0",
});
ai = new GoogleGenAI({});
await client.connect(transport);
});
afterEach(async () => client.close());
test("search_queue is called for simple queue query", async () => {
const interceptor = callToolInterceptor(interceptors);
spyOnClientCallTool(client).spy.mockImplementation((args) =>
interceptor.call(args),
);
const response = await ai.models.generateContent({
model,
contents: `How many queues start with LUCAS?`,
config: {
tools: [mcpToTool(client)],
},
});
console.log("LLM Response", response.text);
const evaluationResults = toolUsageEvaluator(interceptor.toolCallTraces, [
"search_queues",
]);
console.log("Evaluation Results", prettyPrintResults(evaluationResults));
expect(evaluationResults.toolAccuracy).toBe(1);
expect(evaluationResults.extraneousToolsCalled.length).toBe(0);
});
test("Total conversations last week", async () => {
const clientCallToolSpyResult = spyOnClientCallTool(client);
const response = await ai.models.generateContent({
model,
contents: `How many conversations were there between today (${new Date().toISOString()}) and 15 days ago against queues containing TEST in the name?`,
config: {
tools: [mcpToTool(client)],
},
});
console.log("LLM Response", response.text);
console.dir(await clientCallToolSpyResult.tracesFromCallTools(), {
depth: 10,
});
});
},
timeout,
);
