feat: 更新测试客户端以使用 OpenAI 端点并移除 Gemini API 实现

Co-authored-by: aider (vertex_ai/gemini-2.5-pro) <aider@aider.chat>

Author: Dango233

packages/mcp-server/src/bridge/bridge.ts

@@ -129,7 +129,6 @@ export class GcliMcpBridge {
     for (const tool of allTools) {
       this.registerGcliTool(tool);
     }
-    this.registerGeminiApiTool();
   }
 
   private registerGcliTool(tool: GcliTool) {
@@ -149,111 +148,6 @@ export class GcliMcpBridge {
     );
   }
 
-  private registerGeminiApiTool() {
-    this.mcpServer.registerTool(
-      'call_gemini_api',
-      {
-        title: 'Gemini API Proxy',
-        description:
-          "Proxies a request to the Gemini API through the CLI's authenticated client. Allows dynamic provision of tools and a system prompt for this call.",
-        inputSchema: {
-          messages: z
-            .any()
-            .describe(
-              'The conversation history or prompt to send to the Gemini API.',
-            ),
-          tools: z
-            .array(z.any())
-            .optional()
-            .describe(
-              'An array of tool definitions (FunctionDeclarations) to make available to the model for this call.',
-            ),
-          systemInstruction: z
-            .string()
-            .optional()
-            .describe(
-              "A system prompt to guide the model's behavior for this call.",
-            ),
-        },
-      },
-      async (args, { sendNotification, signal }) => {
-        const { messages, tools, systemInstruction } = args as {
-          messages: Content[];
-          tools?: Tool[];
-          systemInstruction?: string;
-        };
-
-        const contentGenerator = this.config
-          .getGeminiClient()
-          .getContentGenerator();
-
-        // 1. Prepare the generation config with dynamic tools and system prompt.
-        const generationConfig: GenerateContentConfig = {
-          tools: tools,
-          systemInstruction: systemInstruction,
-        };
-
-        // The history for the one-shot chat is all messages except the last one.
-        const history = messages.slice(0, -1);
-        // The new prompt is the parts from the last message.
-        const lastMessage = messages[messages.length - 1];
-        const newPrompt = lastMessage?.parts;
-
-        if (!newPrompt) {
-          // This should ideally return a proper JSON-RPC error.
-          // For now, we'll let it proceed, which will likely fail downstream
-          // in sendMessageStream if `newPrompt` is undefined.
-          console.error(
-            `${LOG_PREFIX} ❌ Invalid 'call_gemini_api' arguments: 'messages' array is empty or last message has no parts.`,
-          );
-        }
-
-        // 2. Create a new, stateless GeminiChat instance for this single call.
-        const oneShotChat = new GeminiChat(
-          this.config,
-          contentGenerator,
-          generationConfig, // Pass dynamic config here
-          history, // Start with the provided history
-        );
-
-        // 3. Call sendMessageStream on the new instance.
-        const stream = await oneShotChat.sendMessageStream({
-          message: newPrompt || [], // Pass only the parts of the new message
-        });
-
-        let fullTextResponse = '';
-        for await (const event of stream) {
-          if (signal.aborted) {
-            console.log(`${LOG_PREFIX} 🛑 Request was aborted by the client.`);
-            break;
-          }
-          let chunkText = '';
-          if (event.candidates && event.candidates.length > 0) {
-            const parts = event.candidates[0].content?.parts || [];
-            for (const part of parts) {
-              if (part.text) {
-                chunkText += part.text;
-              }
-            }
-          }
-
-          if (chunkText) {
-            fullTextResponse += chunkText;
-            await sendNotification({
-              method: 'notifications/message',
-              params: { level: 'info', data: `[STREAM_CHUNK]${chunkText}` },
-            });
-          }
-          // Note: Tool call events from the proxied call are not currently forwarded.
-          // This could be a future enhancement if needed.
-        }
-
-        return {
-          content: [{ type: 'text', text: fullTextResponse }],
-        };
-      },
-    );
-  }
 
   private convertJsonSchemaToZod(jsonSchema: any): any {
     // Helper to convert a single JSON schema property to a Zod type.

packages/mcp-server/src/mcp-test-client.ts

@@ -6,6 +6,7 @@ import {
 } from '@modelcontextprotocol/sdk/types.js'; // <--- 引入 Notification 类型
 import { URL } from 'url';
 import { z } from 'zod';
+import OpenAI from 'openai';
 
 // Define the schema for a text content block, as it's not exported by the SDK.
 const TextContentBlockSchema = z.object({
@@ -137,54 +138,60 @@ async function runTestClient() {
           JSON.stringify(tool.inputSchema, null, 2),
         );
       });
-
-      // New test case for call_gemini_api
-      await testGeminiApiCall(client);
     } else {
       logWithPrefix('⚠️ Server returned an empty list of tools.');
     }
   } catch (error) {
     console.error(`${LOG_PREFIX} ❌ Failed to list tools:`, error);
   } finally {
-    logWithPrefix('👋 Closing connection...');
+    logWithPrefix('👋 Closing MCP connection...');
     await client.close();
-    logWithPrefix('🚪 Connection closed. Test finished.');
+    logWithPrefix('🚪 MCP Connection closed.');
   }
+
+  // Now, test the OpenAI endpoint
+  await testOpenAIEndpoint();
+
+  logWithPrefix('✅ Test finished.');
 }
 
-async function testGeminiApiCall(client: Client) {
+async function testOpenAIEndpoint() {
   logWithPrefix('-----------------------------------');
-  logWithPrefix('🚀 Testing "call_gemini_api" tool...');
+  logWithPrefix('🚀 Testing OpenAI compatible endpoint...');
 
-  try {
-    const result = await client.request(
-      {
-        method: 'tools/call',
-        params: {
-          name: 'call_gemini_api',
-          arguments: {
-            messages: [
-              { role: 'user', parts: [{ text: 'Why is the sky blue?' }] },
-            ],
-          },
-        },
-      },
-      z.object({ content: z.array(TextContentBlockSchema) }),
-    );
+  const openai = new OpenAI({
+    baseURL: 'http://localhost:8765/v1',
+    apiKey: 'not-needed', // The API key is not used by our local server
+  });
 
-    logWithPrefix('✅ Successfully received response from call_gemini_api!');
-    const responseText = result.content[0]?.text || '';
-    logWithPrefix('   Response Text:', responseText.substring(0, 100) + '...');
+  try {
+    const stream = await openai.chat.completions.create({
+      model: 'gemini-pro', // This can be any string, it's passed to the transformer
+      messages: [{ role: 'user', content: 'Why is the sky blue?' }],
+      stream: true,
+    });
+
+    let fullResponse = '';
+    logWithPrefix('✅ Stream opened. Receiving response...');
+    for await (const chunk of stream) {
+      const content = chunk.choices[0]?.delta?.content || '';
+      fullResponse += content;
+      process.stdout.write(content);
+    }
+    console.log(''); // Newline after stream
 
-    if (responseText.toLowerCase().includes('scattering')) {
+    if (fullResponse.toLowerCase().includes('scattering')) {
       logWithPrefix('✅ Validation successful: Response contains "scattering".');
     } else {
       console.error(
         `${LOG_PREFIX} ❌ Validation failed: Response did not contain "scattering".`,
       );
     }
   } catch (error) {
-    console.error(`${LOG_PREFIX} ❌ Failed to call call_gemini_api:`, error);
+    console.error(
+      `${LOG_PREFIX} ❌ Failed to call OpenAI endpoint:`,
+      error,
+    );
   }
   logWithPrefix('-----------------------------------');
 }