diff --git a/agent/src/cognition/executeTools.ts b/agent/src/cognition/executeTools.ts
index e69de29..810eb1b 100644
--- a/agent/src/cognition/executeTools.ts
+++ b/agent/src/cognition/executeTools.ts
@@ -0,0 +1,22 @@
+import { genAI, getMcpTool } from '../llm/gemini';
+import { FunctionCallingConfigMode } from '@google/genai';
+
+export async function executeTools(instruction: string): Promise<string> {
+  if (!genAI) throw new Error('LLM not configured');
+  const mcpTool = await getMcpTool();
+  const executionPrompt = `Vas a ejecutar una sola herramienta del MCP seg\xFAn el plan. Explic\xE1 brevemente la acci\xF3n y devuelve s\xF3lo el resultado.`;
+
+  const result = await genAI.models.generateContent({
+    model: 'gemini-pro',
+    contents: [{ role: 'user', parts: [{ text: `${executionPrompt}\n${instruction}` }] }],
+    config: {
+      tools: [mcpTool],
+      toolConfig: {
+        functionCallingConfig: { mode: FunctionCallingConfigMode.ANY },
+      },
+    },
+  });
+
+  const text = (result as any).text || '';
+  return text.trim();
+}
diff --git a/agent/src/cognition/generatePlan.ts b/agent/src/cognition/generatePlan.ts
index e69de29..fde0e69 100644
--- a/agent/src/cognition/generatePlan.ts
+++ b/agent/src/cognition/generatePlan.ts
@@ -0,0 +1,33 @@
+import { genAI, getMcpTool } from '../llm/gemini';
+import { FunctionCallingConfigMode } from '@google/genai';
+import { systemPrompt } from '../systemPrompt';
+import type { Conversation } from '../types';
+
+export async function generatePlan(conversation: Conversation, cognitionPrompt: string): Promise<string> {
+  if (!genAI) throw new Error('LLM not configured');
+
+  const mcpTool = await getMcpTool();
+  const planExecutionPrompt = `Est\xE1s generando el plan de acci\xF3n. Solo deb\xE9s consultar el MCP para listar sus capacidades disponibles y luego describir qu\xE9 pasos seguir. Si ya no hay tareas, inclu\xED la frase \"respuesta final\".`;
+
+  const context = conversation.messages
+    .map(m => {
+      const sender = conversation.participants.find(p => p.id === m.from)?.name || m.from;
+      const content = m.text || `[${m.type}]`;
+      return `${sender}: ${content}`;
+    })
+    .join('\n');
+
+  const prompt = `${systemPrompt}\nConversation:\n${context}\n\nCognition:\n${cognitionPrompt}\n\n${planExecutionPrompt}\nPlan:`;
+  const result = await genAI.models.generateContent({
+    model: 'gemini-pro',
+    contents: [{ role: 'user', parts: [{ text: prompt }] }],
+    config: {
+      tools: [mcpTool],
+      toolConfig: {
+        functionCallingConfig: { mode: FunctionCallingConfigMode.ANY },
+      },
+    },
+  });
+  const text = (result as any).text || '';
+  return text.trim();
+}
diff --git a/agent/src/cognition/index.ts b/agent/src/cognition/index.ts
index 16ddc79..98f55c6 100644
--- a/agent/src/cognition/index.ts
+++ b/agent/src/cognition/index.ts
@@ -1,14 +1,34 @@
-export async function iniciarProcesoCognitivo(){
+import type { Conversation } from '../types';
+import { generatePlan } from './generatePlan';
+import { executeTools } from './executeTools';
+
+interface CognitionArgs {
+  conversation: Conversation;
+}
+
+export async function iniciarProcesoCognitivo({ conversation }: CognitionArgs) {
   try {
-    console.log("Iniciando proceso cognitivo...");
+    console.log('Iniciando proceso cognitivo...');
+    let cognitionPrompt = '';
+    let loopCount = 0;
 
+    while (loopCount < 5) {
+      const plan = await generatePlan(conversation, cognitionPrompt);
+      cognitionPrompt += `\n## Plan\n${plan}\n`;
 
-    //
+      if (/respuesta final/i.test(plan)) {
+        break;
+      }
 
+      const toolResult = await executeTools(plan);
+      cognitionPrompt += `\n## Resultado\n${toolResult}\n`;
+      loopCount += 1;
+    }
 
-
-    console.log("Proceso cognitivo completado.");
+    console.log('Proceso cognitivo completado.');
+    return { text: cognitionPrompt };
   } catch (error) {
-    console.error("Error al iniciar el proceso cognitivo:", error);
+    console.error('Error al iniciar el proceso cognitivo:', error);
+    return { text: 'Error en el proceso cognitivo' };
   }
-}
\ No newline at end of file
+}
diff --git a/agent/src/index.ts b/agent/src/index.ts
index 1248a1c..ed14ae6 100644
--- a/agent/src/index.ts
+++ b/agent/src/index.ts
@@ -51,7 +51,7 @@ app.post('/', async (req, res) => {
 
   try {
     const contents = `${systemPrompt}\nConversation:\n${context}\n`;
-    const result = await iniciarProcesoCognitivo({})
+    const result = await iniciarProcesoCognitivo({ conversation })
     const reply = (result.text || '').trim();
     res.json({ reply });
   } catch (err: any) {