chore: add mcp for generic agent

Author: radugheo

samples/generic-rag-agent/README.md

@@ -1,13 +1,13 @@
-# Generic RAG Agent with Human-in-the-Loop
+# Generic RAG Agent
 
-A generic RAG (Retrieval Augmented Generation) agent that combines multiple knowledge bases with web search capabilities and optional human-in-the-loop approval for query processing.
+A flexible RAG (Retrieval Augmented Generation) agent that combines multiple knowledge bases with web search capabilities and MCP (Model Context Protocol) server integration.
 
 ## Features
 
 - **Multiple Knowledge Base Support**: Query multiple context grounding indexes simultaneously
 - **Web Search Integration**: Real-time information retrieval via Tavily API
+- **MCP Server Integration**: Connect to external MCP servers to access additional tools dynamically
 - **ReAct Agent**: Uses ReAct reasoning to plan and execute queries across multiple tools
-- **Human-in-the-Loop Approval**: Optional human approval before processing queries for enhanced control
 - **Structured Output**: Generates formatted reports with summary, key findings, analysis, and recommendations
 
 ## Prerequisites
@@ -16,6 +16,7 @@ A generic RAG (Retrieval Augmented Generation) agent that combines multiple know
 - UiPath Orchestrator account with Context Grounding service access
 - OpenAI API key (configured in UiPath)
 - (Optional) Tavily API key for web search functionality
+- (Optional) MCP server URL for accessing additional tools
 
 ## Configuration
 
@@ -30,10 +31,22 @@ UIPATH_URL=https://cloud.uipath.com/your_account/your_tenant
 UIPATH_TENANT_ID=your_tenant_id
 UIPATH_ORGANIZATION_ID=your_organization_id
 
-# Optional - API Keys (if using Tavily web search)
+# Optional - Web Search
 TAVILY_API_KEY=your_tavily_api_key
+
+# Optional - MCP Server Integration
+UIPATH_MCP_URL=https://your-mcp-server-url
 ```
 
+**MCP Server Integration:**
+When `UIPATH_MCP_URL` is configured, the agent will:
+1. Connect to the MCP server at startup
+2. Retrieve all available tools from the server
+3. Add them to the agent's toolset alongside knowledge base and web search tools
+4. Keep the MCP session alive during agent execution so tools can make remote calls
+
+The agent gracefully handles MCP connection failures and continues with base tools only.
+
 ### Knowledge Base Configuration
 
 The agent supports multiple knowledge bases configured in [main.py](main.py:21-34):
@@ -80,15 +93,13 @@ The agent accepts the following input parameters in `input.json`:
 ```json
 {
   "query": "Your question here",
-  "add_data_to_index": false,
-  "include_hitl": false
+  "add_data_to_index": false
 }
 ```
 
 **Parameters:**
 - `query` (string): The question or query to process
 - `add_data_to_index` (boolean): Whether to index data before querying (default: false)
-- `include_hitl` (boolean): Whether to require human approval before processing the query (default: false)
 
 ### Running the Agent
 
@@ -101,8 +112,7 @@ Query existing knowledge bases without re-indexing:
 ```json
 {
   "query": "What is our remote work policy?",
-  "add_data_to_index": false,
-  "include_hitl": false
+  "add_data_to_index": false
 }
 ```
 
@@ -113,74 +123,76 @@ Add data to indexes before querying:
 ```json
 {
   "query": "What is our remote work policy?",
-  "add_data_to_index": true,
-  "include_hitl": false
+  "add_data_to_index": true
 }
 ```
 
-#### 3. Human-in-the-Loop Mode
+#### 3. Multi-Tool Query Mode
 
-Require human approval before processing the query:
+Ask questions that leverage multiple tools (knowledge bases, web search, and MCP tools):
 
 ```json
 {
-  "query": "What is our remote work policy?",
-  "add_data_to_index": false,
-  "include_hitl": true
+  "query": "What is the company policy on travel spending? What is the price of GPUs in 2025? What is 5+6+7+8+(6*10)?",
+  "add_data_to_index": false
 }
 ```
 
-When `include_hitl` is set to `true`, the agent will:
-
-1. **Request Approval**: Display the query and ask for human approval:
-   ```
-   Query to be processed: "What is our remote work policy?"
-
-   Is it okay to proceed with this query? (yes/no)
-   ```
-2. **Proceed on Approval**: If "yes", continue with query processing
-3. **Cancel on Rejection**: If "no" (or any other response), stop and return: "Query processing cancelled by user."
+This query will cause the agent to:
+- Query the knowledge base for company policy information
+- Use web search (Tavily) for current GPU pricing
+- Use MCP math tools (if configured) for calculation
 
 ### Workflow Steps
 
 The agent follows this workflow:
 
-1. **Entry Point** ([main.py:158-171](main.py#L158-L171)): Receives query and routing parameters, directs to HITL approval if `include_hitl` is true
-2. **HITL Approval** ([main.py:173-191](main.py#L173-L191)): Requests human approval before proceeding (if `include_hitl` is true)
-3. **Data Indexing** (optional): Adds documents to Context Grounding indexes if `add_data_to_index` is true
-4. **Wait for Ingestion**: Ensures indexes are ready before querying
-5. **Process Query with Agent**: Uses ReAct Agent to gather information from knowledge bases and web search
-6. **Format Final Answer**: Synthesizes and formats the response into a structured report
+1. **Entry Point**: Receives query and routing parameters
+2. **Data Indexing** (optional): Adds documents to Context Grounding indexes if `add_data_to_index` is true
+3. **Wait for Ingestion**: Ensures indexes are ready before querying
+4. **Process Query**:
+   - Initializes base tools (knowledge bases and web search)
+   - Connects to MCP server (if `UIPATH_MCP_URL` is configured) and retrieves MCP tools
+   - Creates ReAct Agent with all available tools
+   - Executes the agent to gather and synthesize information
+   - Keeps MCP session alive during agent execution
+5. **Format Final Answer**: Synthesizes and formats the response into a structured report
 
-## Example Input
+## Example Inputs
 
+### Simple Query
 ```json
 {
-  "query": "What is our remote work policy and what are the latest AI coding tools available in 2025?",
-  "add_data_to_index": false,
-  "include_hitl": true
+  "query": "What is our remote work policy?",
+  "add_data_to_index": false
 }
 ```
 
-## Security Considerations
-
-The human-in-the-loop query approval feature ensures that:
+### Complex Multi-Tool Query
+```json
+{
+  "query": "What is our remote work policy and what are the latest AI coding tools available in 2025? Also calculate 15% of 2500.",
+  "add_data_to_index": false
+}
+```
 
-- **Explicit Consent**: Queries are only processed with explicit human approval when `include_hitl` is enabled
-- **Transparency**: The exact query text is displayed before requesting approval
-- **User Control**: Users can cancel query processing at any time by responding with anything other than "yes"
-- **Audit Trail**: All approval responses are logged for tracking
+This will leverage:
+- **Knowledge bases** for company policy
+- **Web search** for current AI tool information
+- **MCP tools** (if configured) for percentage calculation
 
 ## Customization
 
 ### Adding Custom Tools
 
-Add custom tools by modifying the [generate_context_grounding_query_engine_tools](main.py:102-136) function:
+Add custom tools by modifying the `generate_tools` function in [main.py](main.py):
 
 ```python
-def generate_context_grounding_query_engine_tools(
+def generate_tools(
     response_mode: ResponseMode,
 ) -> list[QueryEngineTool | FunctionTool]:
+    # ... existing tools ...
+
     # Add your custom tools here
     custom_tool = FunctionTool.from_defaults(
         fn=your_custom_function,
@@ -192,21 +204,32 @@ def generate_context_grounding_query_engine_tools(
     return tools
 ```
 
+### Connecting to Different MCP Servers
+
+Simply update the `UIPATH_MCP_URL` in your `.env` file to point to a different MCP server. The agent will automatically connect and load all available tools from that server.
+
 ### Modifying the System Prompts
 
-Update the agent behavior by modifying the system prompts in:
-- ReAct Agent: [main.py:304-310](main.py#L304-L310)
-- Final Answer Formatting: [main.py:329-352](main.py#L329-L352)
+Update the agent behavior by modifying the system prompts in the `process_query` step in [main.py](main.py)
 
 ### Adjusting Knowledge Bases
 
-Modify the `INDEX_CONFIGS` dictionary in [main.py:21-34](main.py#L21-L34) to add, remove, or modify knowledge bases.
+Modify the `INDEX_CONFIGS` dictionary in [main.py](main.py) to add, remove, or modify knowledge bases.
 
 ## Troubleshooting
 
+### MCP Connection Issues
+
+If MCP tools are not loading:
+- Verify `UIPATH_MCP_URL` is correctly set in your `.env` file
+- Check that your `UIPATH_ACCESS_TOKEN` is valid and has not expired
+- Ensure the MCP server is running and accessible
+- Review the console output for MCP connection error messages
+- The agent will continue with base tools only if MCP connection fails
+
 ### Index Ingestion Timeout
 
-- Increase the retry count (line 254) or wait time (line 255) in the `wait_for_index_ingestion` step
+- Increase the retry count or wait time in the `wait_for_index_ingestion` step
 - Check Context Grounding service status in UiPath Orchestrator
 - Verify your documents are properly formatted
 
@@ -216,6 +239,13 @@ Modify the `INDEX_CONFIGS` dictionary in [main.py:21-34](main.py#L21-L34) to add
 - Check that the Tavily API key has sufficient credits
 - Review error messages in the agent logs
 
+### Agent Not Using MCP Tools
+
+The agent uses the ReAct reasoning pattern and decides which tools to use based on the query. If MCP tools aren't being used:
+- Make sure your query requires functionality that the MCP tools provide
+- Verify that MCP tool descriptions are clear and relevant
+- Check the agent's reasoning output (verbose mode) to see why tools were or weren't selected
+
 ### Missing Dependencies
 
 Run the installation command again:
@@ -225,9 +255,5 @@ uv sync
 
 ## Related Samples
 
-- [simple-hitl-agent](../simple-hitl-agent/): Basic HITL implementation with tool-level approval
-- [action-center-hitl-agent](../action-center-hitl-agent/): HITL with UiPath Action Center integration
-
-## License
-
-This sample is provided as-is for demonstration purposes.
+- [simple-remote-mcp-agent](../simple-remote-mcp-agent/): Simple MCP server integration example
+- [context-grounding-retriever-agent](../context-grounding-retriever-agent/): Basic context grounding implementation

samples/generic-rag-agent/input.json

@@ -1,5 +1,5 @@
 {
-  "query": "What is our remote work policy and what are the latest AI coding tools available in 2025?",
+  "query": "What is the company policy on travel spending? What is the price of GPUs in 2025? What is 5+6+7+8+(6*10)?",
   "add_data_to_index": false,
   "include_hitl": false
 }