|
| 1 | +/** |
| 2 | + * @license |
| 3 | + * Copyright 2026 Google LLC |
| 4 | + * SPDX-License-Identifier: Apache-2.0 |
| 5 | + */ |
| 6 | + |
| 7 | +import fs from 'node:fs'; |
| 8 | +import path from 'node:path'; |
| 9 | +import {describe, test} from 'node:test'; |
| 10 | + |
| 11 | +import { |
| 12 | + GoogleGenerativeAI, |
| 13 | + type FunctionDeclaration, |
| 14 | + SchemaType, |
| 15 | +} from '@google/generative-ai'; |
| 16 | +import {Client} from '@modelcontextprotocol/sdk/client/index.js'; |
| 17 | +import {StdioClientTransport} from '@modelcontextprotocol/sdk/client/stdio.js'; |
| 18 | + |
| 19 | +const ROOT_DIR = path.resolve(import.meta.dirname, '..'); |
| 20 | +const SCENARIOS_DIR = path.join(import.meta.dirname, 'eval_scenarios'); |
| 21 | + |
| 22 | +// Define schema for our test scenarios |
| 23 | +export interface CapturedFunctionCall { |
| 24 | + name: string; |
| 25 | + args: Record<string, unknown>; |
| 26 | +} |
| 27 | + |
| 28 | +interface TestScenario { |
| 29 | + prompt: string; |
| 30 | + maxTurns: number; |
| 31 | + expectations?: (calls: CapturedFunctionCall[]) => void; |
| 32 | +} |
| 33 | + |
| 34 | +async function loadScenario(scenarioPath: string): Promise<TestScenario> { |
| 35 | + // Dynamic import of the test scenario |
| 36 | + // We assume the test file exports a 'scenario' object |
| 37 | + const module = await import(scenarioPath); |
| 38 | + if (!module.scenario) { |
| 39 | + throw new Error( |
| 40 | + `Scenario file ${scenarioPath} does not export a 'scenario' object.`, |
| 41 | + ); |
| 42 | + } |
| 43 | + return module.scenario; |
| 44 | +} |
| 45 | + |
| 46 | +async function runSingleScenario( |
| 47 | + scenarioPath: string, |
| 48 | + apiKey: string, |
| 49 | +): Promise<void> { |
| 50 | + const absolutePath = path.resolve(scenarioPath); |
| 51 | + console.log(`\n### Running Scenario: ${absolutePath} ###`); |
| 52 | + |
| 53 | + let client: Client | undefined; |
| 54 | + let transport: StdioClientTransport | undefined; |
| 55 | + |
| 56 | + try { |
| 57 | + const scenario = await loadScenario(absolutePath); |
| 58 | + |
| 59 | + // Path to the compiled MCP server |
| 60 | + const serverPath = path.join(ROOT_DIR, 'build/src/index.js'); |
| 61 | + if (!fs.existsSync(serverPath)) { |
| 62 | + throw new Error( |
| 63 | + `MCP server not found at ${serverPath}. Please run 'npm run build' first.`, |
| 64 | + ); |
| 65 | + } |
| 66 | + |
| 67 | + // Environment variables |
| 68 | + const env: Record<string, string> = {}; |
| 69 | + Object.entries(process.env).forEach(([key, value]) => { |
| 70 | + if (value !== undefined) { |
| 71 | + env[key] = value; |
| 72 | + } |
| 73 | + }); |
| 74 | + |
| 75 | + transport = new StdioClientTransport({ |
| 76 | + command: 'node', |
| 77 | + args: [serverPath], |
| 78 | + env, |
| 79 | + }); |
| 80 | + |
| 81 | + client = new Client( |
| 82 | + {name: 'gemini-eval-client', version: '1.0.0'}, |
| 83 | + {capabilities: {}}, |
| 84 | + ); |
| 85 | + |
| 86 | + await client.connect(transport); |
| 87 | + |
| 88 | + const toolsResult = await client.listTools(); |
| 89 | + const mcpTools = toolsResult.tools; |
| 90 | + |
| 91 | + // Helper to sanitize schema for Gemini |
| 92 | + function isRecord(v: unknown): v is Record<string, unknown> { |
| 93 | + return typeof v === 'object' && v !== null && !Array.isArray(v); |
| 94 | + } |
| 95 | + |
| 96 | + const cleanSchemaRecursive = (schema: unknown): unknown => { |
| 97 | + if (!isRecord(schema)) { |
| 98 | + return schema; |
| 99 | + } |
| 100 | + |
| 101 | + const out: Record<string, unknown> = {}; |
| 102 | + for (const key in schema) { |
| 103 | + if ( |
| 104 | + key === 'default' || |
| 105 | + key === 'additionalProperties' || |
| 106 | + key === 'exclusiveMinimum' |
| 107 | + ) { |
| 108 | + continue; |
| 109 | + } |
| 110 | + |
| 111 | + const value = schema[key]; |
| 112 | + if (Array.isArray(value)) { |
| 113 | + out[key] = value.map(cleanSchemaRecursive); |
| 114 | + } else if (isRecord(value)) { |
| 115 | + out[key] = cleanSchemaRecursive(value); |
| 116 | + } else { |
| 117 | + out[key] = value; |
| 118 | + } |
| 119 | + } |
| 120 | + return out; |
| 121 | + }; |
| 122 | + |
| 123 | + // Convert MCP tools to Gemini function declarations |
| 124 | + const functionDeclarations: FunctionDeclaration[] = mcpTools.map(tool => ({ |
| 125 | + name: tool.name.replace(/-/g, '_').replace(/\./g, '_'), // Sanitize name for Gemini |
| 126 | + description: tool.description?.substring(0, 1024) || '', |
| 127 | + parameters: cleanSchemaRecursive({ |
| 128 | + type: SchemaType.OBJECT, |
| 129 | + properties: |
| 130 | + isRecord(tool.inputSchema) && 'properties' in tool.inputSchema |
| 131 | + ? tool.inputSchema.properties |
| 132 | + : {}, |
| 133 | + required: |
| 134 | + isRecord(tool.inputSchema) && |
| 135 | + 'required' in tool.inputSchema && |
| 136 | + Array.isArray(tool.inputSchema.required) |
| 137 | + ? tool.inputSchema.required |
| 138 | + : [], |
| 139 | + }) as FunctionDeclaration['parameters'], |
| 140 | + })); |
| 141 | + |
| 142 | + // Keep a map of sanitized names to original names for execution |
| 143 | + const contentToolsMap = new Map<string, string>(); |
| 144 | + mcpTools.forEach(tool => { |
| 145 | + const sanitized = tool.name.replace(/-/g, '_').replace(/\./g, '_'); |
| 146 | + contentToolsMap.set(sanitized, tool.name); |
| 147 | + }); |
| 148 | + |
| 149 | + const genAI = new GoogleGenerativeAI(apiKey); |
| 150 | + const model = genAI.getGenerativeModel({ |
| 151 | + model: 'gemini-3-pro-preview', |
| 152 | + tools: [{functionDeclarations}], |
| 153 | + }); |
| 154 | + |
| 155 | + const chat = model.startChat({ |
| 156 | + systemInstruction: { |
| 157 | + role: 'system', |
| 158 | + parts: [{text: `Use available tools.`}], |
| 159 | + }, |
| 160 | + }); |
| 161 | + |
| 162 | + const expectations = scenario.expectations; |
| 163 | + const allCalls: CapturedFunctionCall[] = []; |
| 164 | + |
| 165 | + // Execute turns |
| 166 | + let turnCount = 0; |
| 167 | + console.log(`\n--- Turn 1 (User) ---`); |
| 168 | + console.log(scenario.prompt); |
| 169 | + |
| 170 | + let result = await chat.sendMessage(scenario.prompt); |
| 171 | + let response = result.response; |
| 172 | + |
| 173 | + while (turnCount < scenario.maxTurns) { |
| 174 | + turnCount++; |
| 175 | + console.log(`\n--- Turn ${turnCount} (Model) ---`); |
| 176 | + const text = response.text(); |
| 177 | + if (text) { |
| 178 | + console.log(`Text: ${text}`); |
| 179 | + } |
| 180 | + |
| 181 | + const functionCalls = response.functionCalls(); |
| 182 | + if (functionCalls && functionCalls.length > 0) { |
| 183 | + console.log( |
| 184 | + `Function Calls: ${JSON.stringify(functionCalls, null, 2)}`, |
| 185 | + ); |
| 186 | + |
| 187 | + const functionResponses = []; |
| 188 | + for (const call of functionCalls) { |
| 189 | + const originalName = contentToolsMap.get(call.name); |
| 190 | + if (!originalName) { |
| 191 | + console.error(`Unknown tool called: ${call.name}`); |
| 192 | + functionResponses.push({ |
| 193 | + functionResponse: { |
| 194 | + name: call.name, |
| 195 | + response: {error: `Unknown tool: ${call.name}`}, |
| 196 | + }, |
| 197 | + }); |
| 198 | + continue; |
| 199 | + } |
| 200 | + |
| 201 | + const safeArgs = isRecord(call.args) ? call.args : {}; |
| 202 | + |
| 203 | + console.log( |
| 204 | + `Executing tool: ${originalName} with args: ${JSON.stringify(call.args)}`, |
| 205 | + ); |
| 206 | + |
| 207 | + allCalls.push({ |
| 208 | + name: originalName, |
| 209 | + args: safeArgs, |
| 210 | + }); |
| 211 | + |
| 212 | + try { |
| 213 | + const toolResult = await client.callTool({ |
| 214 | + name: originalName, |
| 215 | + arguments: safeArgs, |
| 216 | + }); |
| 217 | + |
| 218 | + functionResponses.push({ |
| 219 | + functionResponse: { |
| 220 | + name: call.name, |
| 221 | + response: {name: call.name, content: toolResult}, |
| 222 | + }, |
| 223 | + }); |
| 224 | + } catch (e) { |
| 225 | + const errorMessage = e instanceof Error ? e.message : String(e); |
| 226 | + console.error(`Error executing tool ${originalName}:`, e); |
| 227 | + functionResponses.push({ |
| 228 | + functionResponse: { |
| 229 | + name: call.name, |
| 230 | + response: {error: errorMessage}, |
| 231 | + }, |
| 232 | + }); |
| 233 | + } |
| 234 | + } |
| 235 | + |
| 236 | + // Send tool results back |
| 237 | + console.log(`Sending ${functionResponses.length} tool outputs back...`); |
| 238 | + result = await chat.sendMessage(functionResponses); |
| 239 | + response = result.response; |
| 240 | + } else { |
| 241 | + console.log('No tool calls. Interaction finished.'); |
| 242 | + break; |
| 243 | + } |
| 244 | + } |
| 245 | + |
| 246 | + console.log('\nVerifying expectations...'); |
| 247 | + if (expectations) { |
| 248 | + expectations(allCalls); |
| 249 | + } |
| 250 | + } finally { |
| 251 | + if (client) { |
| 252 | + await client.close(); |
| 253 | + } |
| 254 | + if (transport) { |
| 255 | + await transport.close(); |
| 256 | + } |
| 257 | + } |
| 258 | +} |
| 259 | + |
| 260 | +void describe('Gemini Eval Scenarios', () => { |
| 261 | + const apiKey = process.env.GEMINI_API_KEY; |
| 262 | + |
| 263 | + const files = fs.readdirSync(SCENARIOS_DIR).filter(file => { |
| 264 | + return file.endsWith('.ts') || file.endsWith('.js'); |
| 265 | + }); |
| 266 | + |
| 267 | + for (const file of files) { |
| 268 | + void test(file, async () => { |
| 269 | + if (!apiKey) { |
| 270 | + throw new Error('GEMINI_API_KEY environment variable is required.'); |
| 271 | + } |
| 272 | + await runSingleScenario(path.join(SCENARIOS_DIR, file), apiKey); |
| 273 | + }); |
| 274 | + } |
| 275 | +}); |
0 commit comments