ac
Analyze advertising code to identify injection patterns in LLM responses, demonstrating risks of ad middleware in AI systems.
Instructions
ac
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| code | Yes |
Implementation Reference
- src/mcp.ts:15-36 (handler)Core handler function that forwards tool calls (including 'ac') to the remote API by making a POST request with toolName as 'func' header and arguments as body.const calcXiaoBenYangApi = async function (fullArgs: Record<string, any>) { // 发起 POST 请求 let response = await fetch('https://mcp.xiaobenyang.com/api', { method: 'POST', headers: { 'XBY-APIKEY': apiKey, 'func': fullArgs.toolName, 'mcpid': mcpID }, body: new URLSearchParams(fullArgs) }); const apiResult = await response.text(); return { content: [ { type: "text", text: apiResult // 将字符串结果放入 content 中 } ] } as { [x: string]: unknown; content: [{ type: "text"; text: string }] }; };
- src/mcp.ts:39-48 (handler)Wrapper handler registered for each tool, including 'ac', which adds the toolName to arguments and calls the core API forwarder.const handleXiaoBenYangApi = async (args: Record<string, any>, toolName: string) => { // 校验aid是否存在 if (toolName === undefined || toolName === null) { throw new Error("缺少必要参数 'aid'"); } // 合并参数 const fullArgs = {...args, toolName: toolName}; // 调用API return calcXiaoBenYangApi(fullArgs); };
- src/mcp.ts:50-65 (registration)Helper function that registers a tool with the MCP server, called dynamically for each tool including 'ac', specifying the schema and the wrapper handler.const addToolXiaoBenYangApi = function ( name: string, desc: string, params: Record<string, ZodType> ) { server.registerTool( name, { title: name, description: desc, inputSchema: params, } , async (args: Record<string, any>) => handleXiaoBenYangApi(args, name) ) };
- src/mcp.ts:89-132 (registration)Dynamic registration loop that fetches tool list from remote, builds Zod schema for each (including 'ac'), and registers them using addToolXiaoBenYangApi.for (const apiDesc of apiDescList) { let inputSchema = JSON.parse(apiDesc.inputSchema); const zodDict: Record<string, z.ZodTypeAny> = {}; Object.entries(inputSchema.properties).forEach(([name, propConfig]) => { let zodType; let pt = (propConfig as { type: string }).type; switch (pt) { case 'string': zodType = z.string(); break; case 'number': zodType = z.number(); break; case 'boolean': zodType = z.boolean(); break; case 'integer': zodType = z.int32(); break; case 'array': zodType = z.array(z.any()); break; case 'object': zodType = z.object(z.any()); break; default: zodType = z.any(); } if (inputSchema.required?.includes(name)) { zodDict[name] = zodType; } else { zodDict[name] = zodType.optional(); } }); addToolXiaoBenYangApi( apiDesc.name, apiDesc.description ? apiDesc.description : apiDesc.name, zodDict); }
- src/mcp.ts:94-127 (schema)Dynamic schema construction from remote tool descriptions, converting JSON schema types to Zod types for input validation of tools like 'ac'.Object.entries(inputSchema.properties).forEach(([name, propConfig]) => { let zodType; let pt = (propConfig as { type: string }).type; switch (pt) { case 'string': zodType = z.string(); break; case 'number': zodType = z.number(); break; case 'boolean': zodType = z.boolean(); break; case 'integer': zodType = z.int32(); break; case 'array': zodType = z.array(z.any()); break; case 'object': zodType = z.object(z.any()); break; default: zodType = z.any(); } if (inputSchema.required?.includes(name)) { zodDict[name] = zodType; } else { zodDict[name] = zodType.optional(); } });