🐫 CAMEL: The first and the best multi-agent framework. Finding the Scaling Law of Agents. https://www.camel-ai.org
Python agentai-societiesartificial-intelligencecommunicative-aicooperative-aideep-learninglarge-language-modelsmulti-agent-systemsnatural-language-processing
Faiss RAG Agent To MCP Server via ".to_mcp()" method cookbook (#2590)
Divyansh Goyal755f0a828f74
diff --git a/docs/cookbooks/mcp/agent_to_mcp_with_faiss.ipynb b/docs/cookbooks/mcp/agent_to_mcp_with_faiss.ipynbnew file mode 100644index 000000000..81c40ed97--- /dev/null+++ b/docs/cookbooks/mcp/agent_to_mcp_with_faiss.ipynb@@ -0,0 +1,752 @@+{+ "nbformat": 4,+ "nbformat_minor": 0,+ "metadata": {+ "colab": {+ "provenance": []+ },+ "kernelspec": {+ "name": "python3",+ "display_name": "Python 3"+ },+ "language_info": {+ "name": "python"+ }+ },+ "cells": [+ {+ "cell_type": "markdown",+ "source": [+ "# 🐫 **CAMEL Agent to MCP Server Conversion via .to_mcp() for FAISS RAG**\n"+ ],+ "metadata": {+ "id": "fNv8EK3G2zDk"+ }+ },+ {+ "cell_type": "markdown",+ "source": [+ "<div class=\"align-center\">\n",+ " <a href=\"https://www.camel-ai.org/\"><img src=\"https://i.postimg.cc/KzQ5rfBC/button.png\"width=\"150\"></a>\n",+ " <a href=\"https://discord.camel-ai.org\"><img src=\"https://i.postimg.cc/L4wPdG9N/join-2.png\" width=\"150\"></a></a>\n",+ " \n",+ "⭐ <i>Star us on [*Github*](https://github.com/camel-ai/camel), join our [*Discord*](https://discord.camel-ai.org) or follow our [*X*](https://x.com/camelaiorg)\n",+ "</div>"+ ],+ "metadata": {+ "id": "u9C6kQIp3y_C"+ }+ },+ {+ "cell_type": "markdown",+ "source": [],+ "metadata": {+ "id": "FmREQ-bS2YGC"+ }+ },+ {+ "cell_type": "markdown",+ "source": [+ "## Overview\n",+ "This cookbook demonstrates how to convert a CAMEL RAG agent into an MCP server using the .to_mcp() method, allowing other MCP clients to use your agent as a tool.\n",+ "## What You'll Learn\n",+ "- How to create a self-contained RAG agent with embedded FAISS tools\n",+ "- How to convert any CAMEL agent to an MCP server using `.to_mcp()`\n",+ "- How to test the converted MCP server with a client\n",+ "\n",+ "\n",+ "## 🎯 Key Concept: The .to_mcp() Method\n",+ "The `.to_mcp()` method converts any CAMEL agent into an MCP server, making it available as a tool for other MCP clients.\n",+ "\n",+ "\n",+ "\n",+ "\n"+ ],+ "metadata": {+ "id": "HM_n50_G33-R"+ }+ },+ {+ "cell_type": "code",+ "source": [+ "# Simple example from CAMEL documentation\n",+ "from camel.agents import ChatAgent\n",+ "\n",+ "agent = ChatAgent(model=\"gpt-4o-mini\")\n",+ "mcp_server = agent.to_mcp(name=\"demo\", description=\"A demo agent\")\n",+ "mcp_server.run(transport=\"stdio\")"+ ],+ "metadata": {+ "id": "WKbAedpuT7ld"+ },+ "execution_count": null,+ "outputs": []+ },+ {+ "cell_type": "markdown",+ "source": [+ "# File 1: RAG Agent Configuration (faiss_rag_agent_config.py)\n",+ "\n",+ "## Purpose: This file creates a self-contained RAG agent with embedded FAISS vector search tools.\n",+ "\n",+ "## What it does:\n",+ "\n",+ "Defines three core FAISS tools: create_faiss_index, query_faiss, and get_index_stats\n",+ "\n",+ "- Implements simple text chunking and vector embeddings using numpy\n",+ "- Creates a ChatAgent with Gemini model and embedded tools\n",+ "- Provides a complete RAG system without external dependencies\n",+ "- Self-contained: No external MCP server dependencies\n",+ "- FAISS integration: Uses Facebook AI Similarity Search for vector operations\n",+ "- Simple embeddings: Uses deterministic random embeddings for demonstration\n",+ "- Tool composition: Wraps functions as FunctionTool objects for CAMEL\n"+ ],+ "metadata": {+ "id": "p3rs66Il4aXz"+ }+ },+ {+ "cell_type": "code",+ "source": [+ "import logging\n",+ "import os\n",+ "from pathlib import Path\n",+ "from typing import List, Dict, Any\n",+ "import json\n",+ "\n",+ "from camel.agents import ChatAgent\n",+ "from camel.messages import BaseMessage\n",+ "from camel.models import ModelFactory\n",+ "from camel.types import ModelPlatformType, RoleType\n",+ "from camel.toolkits import FunctionTool\n",+ "from camel.storages import FaissStorage\n",+ "from camel.storages.vectordb_storages.base import VectorRecord, VectorDBQuery\n",+ "from camel.types.enums import VectorDistance\n",+ "import numpy as np\n",+ "\n",+ "# Prevent logging since MCP needs to use stdout\n",+ "root_logger = logging.getLogger()\n",+ "root_logger.handlers = []\n",+ "root_logger.addHandler(logging.NullHandler())\n",+ "\n",+ "# Global storage instance\n",+ "faiss_storage = None\n",+ "\n",+ "def initialize_faiss_storage():\n",+ " \"\"\"Initialize CAMEL's FaissStorage with professional settings\"\"\"\n",+ " global faiss_storage\n",+ "\n",+ " storage_path = Path.cwd() / \"rag_storage\"\n",+ " storage_path.mkdir(exist_ok=True)\n",+ "\n",+ " faiss_storage = FaissStorage(\n",+ " vector_dim=1536, # OpenAI embedding dimension\n",+ " index_type='HNSW', # Fast with high recall\n",+ " collection_name='rag_documents',\n",+ " storage_path=str(storage_path),\n",+ " distance=VectorDistance.COSINE,\n",+ " m=16, # HNSW connections per node\n",+ " )\n",+ "\n",+ " # Load existing index if available\n",+ " try:\n",+ " faiss_storage.load()\n",+ " except:\n",+ " pass # No existing index to load\n",+ "\n",+ " return faiss_storage\n",+ "\n",+ "\n",+ "def create_rag_agent():\n",+ " \"\"\"Create a self-contained RAG agent with CAMEL's FaissStorage\"\"\"\n",+ "\n",+ " # Initialize storage\n",+ " initialize_faiss_storage()\n",+ "\n",+ " # Create Gemini model\n",+ " model = ModelFactory.create(\n",+ " model_platform=ModelPlatformType.GEMINI,\n",+ " model_type=\"gemini-2.5-flash-preview-04-17\",\n",+ " api_key=os.getenv(\"GOOGLE_API_KEY\"),\n",+ " model_config_dict={\n",+ " \"temperature\": 0.2,\n",+ " }\n",+ " )\n",+ "\n",+ " # No tools needed since .to_mcp() doesn't expose them\n",+ " tools = []\n",+ "\n",+ " print(f\"DEBUG: Created {len(tools)} tools:\")\n",+ " for i, tool in enumerate(tools):\n",+ " print(f\" Tool {i+1}: {tool.func.__name__}\")\n",+ "\n",+ " # Enhanced system message that tells the agent it HAS these tools\n",+ " system_content = (\n",+ " \"You are a professional RAG assistant with access to CAMEL's FaissStorage for vector database operations. \"\n",+ " \"You have the following tools available:\\n\"\n",+ " \"1. create_faiss_index(documents, chunk_size) - Create a persistent FAISS index\\n\"\n",+ " \"2. query_faiss(query, top_k) - Search for semantically similar content\\n\"\n",+ " \"3. get_index_stats() - Get detailed statistics about the vector storage\\n\"\n",+ " \"\\nYou MUST use these tools when users ask you to create indexes or search for information. \"\n",+ " \"Do not claim you cannot perform these operations - you have the tools to do them!\"\n",+ " )\n",+ "\n",+ " system_message = BaseMessage(\n",+ " role_name=\"Professional RAG Assistant\",\n",+ " role_type=RoleType.ASSISTANT,\n",+ " meta_dict={\"task\": \"Production FAISS RAG Operations\"},\n",+ " content=system_content\n",+ " )\n",+ "\n",+ " # Create the agent (conversational interface only)\n",+ " rag_agent = ChatAgent(\n",+ " system_message=system_message,\n",+ " model=model,\n",+ " tools=tools # This should attach the tools\n",+ " )\n",+ "\n",+ " # Verify tools are attached - Fixed debug code\n",+ " print(f\"DEBUG: Agent created\")\n",+ " if hasattr(rag_agent, '_internal_tools'):\n",+ " print(f\"DEBUG: Agent has {len(rag_agent._internal_tools)} internal tools\")\n",+ " # Fix: _internal_tools contains tool names (strings), not objects\n",+ " for i, tool_name in enumerate(rag_agent._internal_tools):\n",+ " print(f\" Internal Tool {i+1}: {tool_name}\")\n",+ " else:\n",+ " print(\"DEBUG: Agent has no _internal_tools attribute\")\n",+ "\n",+ " if hasattr(rag_agent, 'tool_dict'):\n",+ " print(f\"DEBUG: Agent tool_dict: {list(rag_agent.tool_dict.keys())}\")\n",+ " # Also show the actual tool objects\n",+ " for tool_name, tool_obj in rag_agent.tool_dict.items():\n",+ " print(f\" {tool_name}: {tool_obj.func.__name__}\")\n",+ "\n",+ " return rag_agent\n",+ "\n"+ ],+ "metadata": {+ "id": "kGjDLuW04en8"+ },+ "execution_count": null,+ "outputs": []+ },+ {+ "cell_type": "markdown",+ "source": [+ "# File 2: MCP Server Conversion (faiss_rag_agent_mcp_server.py)\n",+ "## Purpose: This is the main file that demonstrates the agent-to-MCP conversion.\n",+ "\n",+ "## What it does:\n",+ "- Imports the RAG agent from the configuration file\n",+ "- Uses the .to_mcp() method to convert the agent into an MCP server\n",+ "- Runs the server with stdio transport for local communication\n",+ "\n",+ "## Critical Points:\n",+ "- Must be synchronous (no async/await) - let .to_mcp() handle event loops\n",+ "- Requires GOOGLE_API_KEY environment variable\n"+ ],+ "metadata": {+ "id": "oAL2f5cM6cur"+ }+ },+ {+ "cell_type": "code",+ "source": [+ "import os\n",+ "import json\n",+ "import numpy as np\n",+ "from typing import List\n",+ "from pathlib import Path\n",+ "\n",+ "from faiss_rag_agent_config import (\n",+ " create_rag_agent,\n",+ " initialize_faiss_storage,\n",+ " faiss_storage # Import the global storage\n",+ ")\n",+ "from camel.storages.vectordb_storages.base import VectorRecord, VectorDBQuery\n",+ "\n",+ "def main():\n",+ " \"\"\"Initialize and run the RAG agent as an MCP server with hybrid approach\"\"\"\n",+ "\n",+ " # Ensure API key is set\n",+ " if not os.getenv(\"GOOGLE_API_KEY\"):\n",+ " print(\"Error: GOOGLE_API_KEY environment variable not set\")\n",+ " return\n",+ "\n",+ " print(\"DEBUG: Creating RAG agent...\")\n",+ " # Create the self-contained RAG agent\n",+ " rag_agent = create_rag_agent()\n",+ "\n",+ " # Ensure storage is initialized at the module level\n",+ " print(\"DEBUG: Ensuring FAISS storage is initialized...\")\n",+ " storage_instance = initialize_faiss_storage()\n",+ "\n",+ " print(\"DEBUG: Converting agent to MCP server...\")\n",+ "\n",+ " try:\n",+ " # Step 1: Convert ChatAgent to MCP server (for conversational capabilities)\n",+ " mcp_server = rag_agent.to_mcp(\n",+ " name=\"RAG-Agent-MCP\",\n",+ " description=\"A self-contained RAG assistant with FAISS vector search capabilities\"\n",+ " )\n",+ " print(\"DEBUG: .to_mcp() conversion successful!\")\n",+ "\n",+ " # Step 2: Add custom FAISS tools to the same MCP server instance\n",+ " print(\"DEBUG: Adding custom FAISS tools to MCP server...\")\n",+ "\n",+ " @mcp_server.tool()\n",+ " async def create_faiss_index(documents: List[str], chunk_size: int = 1000) -> str:\n",+ " \"\"\"Create a FAISS index from document texts using CAMEL's storage.\"\"\"\n",+ " # Import here to ensure we get the latest global state\n",+ " from faiss_rag_agent_config import faiss_storage, initialize_faiss_storage\n",+ "\n",+ " try:\n",+ " print(f\"DEBUG: create_faiss_index called with {len(documents)} documents\")\n",+ "\n",+ " # Ensure storage is available\n",+ " current_storage = faiss_storage\n",+ " if current_storage is None:\n",+ " print(\"DEBUG: Storage is None, initializing...\")\n",+ " current_storage = initialize_faiss_storage()\n",+ "\n",+ " print(f\"DEBUG: Storage status before clear: {current_storage.status().vector_count} vectors\")\n",+ " current_storage.clear()\n",+ " print(\"DEBUG: Storage cleared successfully\")\n",+ "\n",+ " all_records = []\n",+ " chunk_id = 0\n",+ "\n",+ " for doc_idx, doc_text in enumerate(documents):\n",+ " words = doc_text.split()\n",+ "\n",+ " for i in range(0, len(words), max(1, chunk_size // 10)):\n",+ " chunk_words = words[i:i + max(1, chunk_size // 10)]\n",+ " chunk_text = ' '.join(chunk_words)\n",+ "\n",+ " if chunk_text.strip():\n",+ " # Create simple embedding\n",+ " np.random.seed(hash(chunk_text) % (2**32))\n",+ " embedding = np.random.normal(0, 1, 1536).astype(np.float32)\n",+ "\n",+ " record = VectorRecord(\n",+ " id=f\"chunk_{chunk_id}\",\n",+ " vector=embedding,\n",+ " payload={\n",+ " \"text\": chunk_text,\n",+ " \"source_id\": doc_idx,\n",+ " \"chunk_id\": chunk_id,\n",+ " \"metadata\": {\n",+ " \"document_index\": doc_idx,\n",+ " \"chunk_index\": chunk_id\n",+ " }\n",+ " }\n",+ " )\n",+ "\n",+ " all_records.append(record)\n",+ " chunk_id += 1\n",+ "\n",+ " if all_records:\n",+ " print(f\"DEBUG: Adding {len(all_records)} records to storage\")\n",+ " current_storage.add(all_records)\n",+ "\n",+ " # Verify the addition\n",+ " final_status = current_storage.status()\n",+ " print(f\"DEBUG: Final storage status: {final_status.vector_count} vectors\")\n",+ "\n",+ " return json.dumps({\n",+ " \"status\": \"success\",\n",+ " \"message\": f\"Indexed {len(all_records)} chunks from {len(documents)} documents\",\n",+ " \"total_chunks\": len(all_records),\n",+ " \"final_vector_count\": final_status.vector_count\n",+ " })\n",+ " else:\n",+ " return json.dumps({\"error\": \"No text chunks were created\"})\n",+ "\n",+ " except Exception as e:\n",+ " import traceback\n",+ " error_details = traceback.format_exc()\n",+ " print(f\"DEBUG: Error in create_faiss_index: {error_details}\")\n",+ " return json.dumps({\"error\": f\"Error creating FAISS index: {str(e)}\", \"details\": error_details})\n",+ "\n",+ " @mcp_server.tool()\n",+ " async def query_faiss(query: str, top_k: int = 3) -> str:\n",+ " \"\"\"Query the FAISS index using CAMEL's storage.\"\"\"\n",+ " # Import here to ensure we get the latest global state\n",+ " from faiss_rag_agent_config import faiss_storage, initialize_faiss_storage\n",+ "\n",+ " try:\n",+ " print(f\"DEBUG: query_faiss called with query: '{query}', top_k: {top_k}\")\n",+ "\n",+ " # Ensure storage is available\n",+ " current_storage = faiss_storage\n",+ " if current_storage is None:\n",+ " print(\"DEBUG: Storage is None, initializing...\")\n",+ " current_storage = initialize_faiss_storage()\n",+ "\n",+ " status = current_storage.status()\n",+ " print(f\"DEBUG: Storage has {status.vector_count} vectors\")\n",+ "\n",+ " if status.vector_count == 0:\n",+ " return json.dumps({\"error\": \"No documents have been indexed yet\"})\n",+ "\n",+ " # Create query embedding\n",+ " np.random.seed(hash(query) % (2**32))\n",+ " query_embedding = np.random.normal(0, 1, 1536).astype(np.float32)\n",+ "\n",+ " db_query = VectorDBQuery(\n",+ " query_vector=query_embedding,\n",+ " top_k=min(top_k, status.vector_count)\n",+ " )\n",+ "\n",+ " results = current_storage.query(db_query)\n",+ "\n",+ " formatted_results = []\n",+ " for result in results:\n",+ " formatted_results.append({\n",+ " \"id\": result.record.id,\n",+ " \"text\": result.record.payload.get(\"text\", \"\"),\n",+ " \"metadata\": result.record.payload.get(\"metadata\", {}),\n",+ " \"score\": float(result.similarity)\n",+ " })\n",+ "\n",+ " return json.dumps({\n",+ " \"results\": formatted_results,\n",+ " \"query_info\": {\n",+ " \"total_results\": len(formatted_results),\n",+ " \"search_type\": \"semantic_similarity\"\n",+ " }\n",+ " })\n",+ "\n",+ " except Exception as e:\n",+ " import traceback\n",+ " error_details = traceback.format_exc()\n",+ " print(f\"DEBUG: Error in query_faiss: {error_details}\")\n",+ " return json.dumps({\"error\": f\"Error querying FAISS index: {str(e)}\", \"details\": error_details})\n",+ "\n",+ " @mcp_server.tool()\n",+ " async def get_index_stats() -> str:\n",+ " \"\"\"Get statistics about the FAISS storage.\"\"\"\n",+ " # Import here to ensure we get the latest global state\n",+ " from faiss_rag_agent_config import faiss_storage, initialize_faiss_storage\n",+ "\n",+ " try:\n",+ " print(\"DEBUG: get_index_stats called\")\n",+ "\n",+ " # Ensure storage is available\n",+ " current_storage = faiss_storage\n",+ " if current_storage is None:\n",+ " print(\"DEBUG: Storage is None, initializing...\")\n",+ " current_storage = initialize_faiss_storage()\n",+ "\n",+ " status = current_storage.status()\n",+ "\n",+ " stats = {\n",+ " \"vector_count\": status.vector_count,\n",+ " \"vector_dimension\": 1536,\n",+ " \"index_type\": \"HNSW\",\n",+ " \"distance_metric\": \"cosine\",\n",+ " \"collection_name\": current_storage.collection_name,\n",+ " \"persistent_storage\": current_storage.storage_path is not None,\n",+ " \"storage_path\": str(current_storage.storage_path) if current_storage.storage_path else \"in-memory\"\n",+ " }\n",+ "\n",+ " return json.dumps(stats, indent=2)\n",+ "\n",+ " except Exception as e:\n",+ " import traceback\n",+ " error_details = traceback.format_exc()\n",+ " print(f\"DEBUG: Error in get_index_stats: {error_details}\")\n",+ " return json.dumps({\"error\": f\"Error getting index stats: {str(e)}\", \"details\": error_details})\n",+ "\n",+ " print(\"DEBUG: Custom FAISS tools added successfully!\")\n",+ " print(\"DEBUG: Starting hybrid MCP server...\")\n",+ "\n",+ " # Run the MCP server with both conversational and tool capabilities\n",+ " mcp_server.run(transport=\"stdio\")\n",+ "\n",+ " except Exception as e:\n",+ " import traceback\n",+ " error_details = traceback.format_exc()\n",+ " print(f\"DEBUG: Hybrid MCP setup failed: {e}\")\n",+ " print(f\"DEBUG: Error details: {error_details}\")\n",+ " return\n",+ "\n",+ "if __name__ == \"__main__\":\n",+ " main()"+ ],+ "metadata": {+ "id": "RD4dhcOb4qWo"+ },+ "execution_count": null,+ "outputs": []+ },+ {+ "cell_type": "markdown",+ "source": [+ "# File 3: Test Client (test_faiss_rag_mcp_client.py)\n",+ "## Purpose: Tests the converted MCP server by connecting to it as a client.\n",+ "\n",+ "## What it does:\n",+ "\n",+ "- Creates an MCP configuration that points to your server script\n",+ "- Connects to the MCP server using MCPToolkit\n",+ "- Creates a client agent that can use the server's tools\n",+ "- Provides an interactive interface to test RAG functionality\n",+ "\n"+ ],+ "metadata": {+ "id": "u0a_VY1A67Kc"+ }+ },+ {+ "cell_type": "code",+ "source": [+ "import asyncio\n",+ "import json\n",+ "import os\n",+ "import sys\n",+ "from getpass import getpass\n",+ "from pathlib import Path\n",+ "\n",+ "from camel.agents import ChatAgent\n",+ "from camel.logger import get_logger\n",+ "from camel.messages import BaseMessage\n",+ "from camel.models import ModelFactory\n",+ "from camel.toolkits import MCPToolkit\n",+ "from camel.types import ModelPlatformType, RoleType\n",+ "\n",+ "logger = get_logger(__name__)\n",+ "\n",+ "async def main():\n",+ " \"\"\"Test the self-contained RAG MCP server\"\"\"\n",+ "\n",+ " # Configuration for connecting to our RAG MCP server\n",+ " rag_mcp_config = {\n",+ " \"mcpServers\": {\n",+ " \"rag_agent_server\": {\n",+ " \"type\": \"script\",\n",+ " \"command\": \"python\",\n",+ " \"args\": [\"faiss_rag_agent_mcp_server.py\"],\n",+ " \"transport\": \"stdio\",\n",+ " \"env\": {\n",+ " \"GOOGLE_API_KEY\": os.getenv(\"GOOGLE_API_KEY\") or getpass('Enter Gemini API key for server: ')\n",+ " }\n",+ " }\n",+ " }\n",+ " }\n",+ "\n",+ " # Write config file\n",+ " config_path = Path.cwd() / \"rag_mcp_client_config.json\"\n",+ " with open(config_path, 'w') as f:\n",+ " json.dump(rag_mcp_config, f, indent=2)\n",+ "\n",+ " # Get API key for client model\n",+ " # gemini_api_key = getpass('Enter your Gemini API key for client: ')\n",+ " # if not gemini_api_key:\n",+ " # logger.error(\"API key is required to proceed.\")\n",+ " # return\n",+ " # os.environ[\"GOOGLE_API_KEY\"] = gemini_api_key\n",+ " gemini_api_key = os.getenv(\"GOOGLE_API_KEY\") or getpass('Enter Gemini API key for client: ')\n",+ "\n",+ " # Connect to the RAG MCP server\n",+ " logger.info(\"Connecting to self-contained RAG MCP server...\")\n",+ " mcp_toolkit = MCPToolkit(config_path=str(config_path))\n",+ "\n",+ " await mcp_toolkit.connect()\n",+ " tools = mcp_toolkit.get_tools()\n",+ "\n",+ " # Debugging:-------------------\n",+ " # Add debug before the existing debug section:\n",+ " print(f\"\\n=== DEBUG: MCP Client received {len(tools)} tools ===\")\n",+ " if tools:\n",+ " for i, tool in enumerate(tools):\n",+ " print(f\"Tool {i+1}:\")\n",+ " print(f\" Name: {getattr(tool, 'name', 'Unknown')}\")\n",+ " if hasattr(tool, 'func'):\n",+ " print(f\" Function: {tool.func.__name__}\")\n",+ " print(f\" Doc: {tool.func.__doc__[:100] if tool.func.__doc__ else 'No doc'}...\")\n",+ " else:\n",+ " print(f\" Type: {type(tool)}\")\n",+ " print(f\" Attributes: {[attr for attr in dir(tool) if not attr.startswith('_')]}\")\n",+ " else:\n",+ " print(\" No tools received from MCP server!\")\n",+ " print(\"=\" * 50)\n",+ " # Debugging--------------------\n",+ "\n",+ " if not tools:\n",+ " logger.error(\"No tools found from the RAG MCP server\")\n",+ " return\n",+ "\n",+ " logger.info(f\"Found {len(tools)} tools from RAG MCP server\")\n",+ "\n",+ " try:\n",+ " # Create a client model to interact with the RAG MCP server\n",+ " model = ModelFactory.create(\n",+ " model_platform=ModelPlatformType.GEMINI,\n",+ " model_type=\"gemini-2.5-flash-preview-04-17\",\n",+ " api_key=gemini_api_key,\n",+ " model_config_dict={\n",+ " \"temperature\": 0.2,\n",+ " # \"max_output_tokens\": 1024,\n",+ " }\n",+ " )\n",+ "\n",+ " system_message = BaseMessage(\n",+ " role_name=\"RAG Client\",\n",+ " role_type=RoleType.ASSISTANT,\n",+ " meta_dict={\"task\": \"RAG MCP Client\"},\n",+ " content=\"You are a client that uses a self-contained RAG MCP server for document search and retrieval tasks.\"\n",+ " )\n",+ "\n",+ " # Create client agent with RAG MCP tools\n",+ " client_agent = ChatAgent(\n",+ " system_message=system_message,\n",+ " model=model,\n",+ " tools=tools\n",+ " )\n",+ " client_agent.reset()\n",+ "\n",+ " # Debugging --------------------------\n",+ "\n",+ " # Add direct tool testing:\n",+ " print(\"\\n=== DEBUG: Testing direct tool access ===\")\n",+ " if tools:\n",+ " try:\n",+ " # Try to call get_index_stats directly\n",+ " stats_tool = None\n",+ " for tool in tools:\n",+ " if hasattr(tool, 'func') and 'stats' in tool.func.__name__:\n",+ " stats_tool = tool\n",+ " break\n",+ "\n",+ " if stats_tool:\n",+ " print(\"Found stats tool, testing direct call...\")\n",+ " result = stats_tool.func()\n",+ " print(f\"Direct tool call result: {result}\")\n",+ " else:\n",+ " print(\"Could not find stats tool\")\n",+ " except Exception as e:\n",+ " print(f\"Direct tool call failed: {e}\")\n",+ " else:\n",+ " print(\"No tools to test directly\")\n",+ " print(\"=\" * 50)\n",+ "# DEBUGGING ----------------------------------\n",+ "\n",+ " print(\"\\n=== Self-Contained RAG Agent as MCP Server Test ===\\n\")\n",+ " print(\"Testing CAMEL RAG agent (with embedded FAISS) exported as MCP server using .to_mcp()...\")\n",+ " print(\"Type 'exit' to quit\\n\")\n",+ "\n",+ " # Test the RAG MCP server\n",+ " test_message = \"Hello! Please create an index with documents about AI, machine learning, and FAISS, then search for information about vector databases.\"\n",+ " print(f\"Test Query: {test_message}\\n\")\n",+ "\n",+ " response = await client_agent.astep(test_message)\n",+ " if response and response.msgs:\n",+ " print(f\"RAG MCP Response: {response.msgs[0].content}\\n\")\n",+ "\n",+ " # Interactive loop\n",+ " while True:\n",+ " user_input = input(\"You: \")\n",+ " if user_input.lower() == 'exit':\n",+ " break\n",+ "\n",+ " response = await client_agent.astep(user_input)\n",+ "\n",+ " if response and response.msgs:\n",+ " agent_reply = response.msgs[0].content\n",+ " print(f\"RAG MCP Agent: {agent_reply}\\n\")\n",+ " else:\n",+ " print(\"No response received.\\n\")\n",+ "\n",+ " except Exception as e:\n",+ " logger.error(f\"An error occurred: {str(e)}\")\n",+ " print(f\"\\nError: {str(e)}\")\n",+ " finally:\n",+ " try:\n",+ " await mcp_toolkit.disconnect()\n",+ " except Exception as cleanup_error:\n",+ " logger.warning(f\"Error during cleanup: {cleanup_error}\")\n",+ "\n",+ "if __name__ == \"__main__\":\n",+ " if sys.platform == \"win32\" and sys.version_info >= (3, 8):\n",+ " asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())\n",+ " asyncio.run(main())"+ ],+ "metadata": {+ "id": "P9ZZ8Pwn41ck"+ },+ "execution_count": null,+ "outputs": []+ },+ {+ "cell_type": "markdown",+ "source": [+ ""+ ],+ "metadata": {+ "id": "mX4Pv4Ha7YC3"+ }+ },+ {+ "cell_type": "markdown",+ "source": [+ ""+ ],+ "metadata": {+ "id": "GwSDg3XHF1Ve"+ }+ },+ {+ "cell_type": "markdown",+ "source": [+ "# Running the Example\n",+ "## Step 1: Set Up Environment\n",+ "```python\n",+ "export GOOGLE_API_KEY=\"your-gemini-api-key-here\"\n",+ "```\n",+ "## Step 2: Test the MCP Server (Optional)\n",+ "This starts the server and waits for connections. If it runs without errors, the conversion worked!\n",+ "```python\n",+ "python faiss_rag_agent_mcp_server.py\n",+ "```\n",+ "## Step 3: Run the Client Test\n",+ "```python\n",+ "python test_faiss_rag_mcp_client.py\n",+ "```\n"+ ],+ "metadata": {+ "id": "TzY2LZq77nLb"+ }+ },+ {+ "cell_type": "markdown",+ "source": [+ "## Key Takeaways\n",+ "\n",+ "### ✅ What We Accomplished\n",+ "\n",+ "1. **Self-Contained Agent**: Created a RAG agent with embedded FAISS tools\n",+ "2. **Agent → MCP Conversion**: Used `.to_mcp()` to convert the agent to an MCP server\n",+ "3. **Cross-Client Compatibility**: The MCP server works with any MCP client\n",+ "4. **Tool Composition**: Other agents can now use our RAG capabilities as tools\n"+ ],+ "metadata": {+ "id": "k8ovcxfU5UBu"+ }+ },+ {+ "cell_type": "markdown",+ "source": [+ "<div class=\"align-center\">\n",+ " <a href=\"https://www.camel-ai.org/\"><img src=\"https://i.postimg.cc/KzQ5rfBC/button.png\"width=\"150\"></a>\n",+ " <a href=\"https://discord.camel-ai.org\"><img src=\"https://i.postimg.cc/L4wPdG9N/join-2.png\" width=\"150\"></a></a>\n",+ " \n",+ "⭐ <i>Star us on [*Github*](https://github.com/camel-ai/camel), join our [*Discord*](https://discord.camel-ai.org) or follow our [*X*](https://x.com/camelaiorg)\n",+ "</div>"+ ],+ "metadata": {+ "id": "LX-fx2OP8Z7Y"+ }+ }+ ]+}\ No newline at end of file