IBM i MCP Server

--- title: "Getting Started with Agno" description: "Introduction to building IBM i agents with the Agno framework and FilteredMCPTools." --- [Agno](https://docs.agno.com/introduction) is a modern Python framework for building AI agents. This guide introduces the core concepts for building IBM i agents with Agno and FilteredMCPTools. ## What is Agno? Agno provides: - **Agent Framework** - Build AI agents with natural language interfaces - **Tool Integration** - Connect agents to external tools via MCP protocol - **Persistent Memory** - SQLite-backed conversation and knowledge storage - **AgentOS** - Deploy and manage multiple specialized agents - **Evaluation Framework** - Test agent reliability systematically ## Core Concepts ### FilteredMCPTools The key innovation for IBM i agents is **FilteredMCPTools** - a specialized class that filters MCP tools by annotations: ```python from ibmi_agents.tools.filtered_mcp_tools import FilteredMCPTools # Agent sees only performance tools (8 of 45+ total) performance_tools = FilteredMCPTools( url="http://localhost:3010/mcp", annotation_filters={"toolsets": ["performance"]} ) ``` **Why filter tools?** - **Faster decisions** - Smaller tool sets = quicker agent reasoning - **Higher accuracy** - Agent chooses from relevant tools only - **Better security** - Limit access to necessary operations - **Improved performance** - Reduced token usage <Card title="FilteredMCPTools Deep Dive" icon="filter" href="/agents/agno/filtered-mcp-tools"> Learn about annotation-based filtering, available toolsets, and advanced patterns </Card> ### Specialized Agents Instead of one agent with all tools, create focused agents per domain: ```python from agno.agent import Agent from agno.models.openai import OpenAIChat # Performance monitoring agent performance_agent = Agent( name="IBM i Performance Monitor", model=OpenAIChat(id="gpt-4o"), tools=[performance_tools], instructions=[ "You are an IBM i performance monitoring specialist.", "Analyze system metrics and identify bottlenecks.", "Explain technical findings in business terms." ] ) ``` **Benefits of specialization:** - Each agent masters specific domain - Clear separation of concerns - Easier to test and maintain - Can run multiple agents in parallel --- ## Quick Start ### Prerequisites Before building agents: 1. **MCP Server Running** - Complete the [Quick Start Guide](/quickstart) 2. **Python 3.8+** installed 3. **OpenAI API Key** from [platform.openai.com](https://platform.openai.com) Includes: - **Performance Agent** - System monitoring (performance toolset) - **Discovery Agent** - Service discovery (sysadmin_discovery toolset) - **Browse Agent** - Detailed exploration (sysadmin_browse toolset) - **Search Agent** - Find specific services (sysadmin_search toolset) <Card title="Example Documentation" icon="github" href="https://github.com/IBM/ibmi-mcp-server/tree/main/agents/frameworks/agno"> **See the Agno examples README** for detailed setup instructions, configuration options, and usage examples </Card> --- ## Building Your First Agent ### Basic Agent Structure ```python import os from agno.agent import Agent from agno.models.openai import OpenAIChat from ibmi_agents.tools.filtered_mcp_tools import FilteredMCPTools # 1. Configure filtered tools tools = FilteredMCPTools( url=os.getenv("MCP_SERVER_URL", "http://localhost:3010/mcp"), transport="streamable-http", annotation_filters={"toolsets": ["performance"]} ) # 2. Create specialized agent agent = Agent( name="Performance Analyst", model=OpenAIChat(id="gpt-4o"), tools=[tools], instructions=[ "You are an IBM i performance analyst.", "Identify performance bottlenecks and recommend solutions.", "Explain metrics in business terms." ] ) # 3. Use the agent if __name__ == "__main__": agent.print_response("Check system performance") ``` ### Key Components <AccordionGroup> <Accordion title="FilteredMCPTools" icon="filter"> Filters available tools by annotations: ```python FilteredMCPTools( url="http://localhost:3010/mcp", annotation_filters={ "toolsets": ["performance"], # Only performance tools "readOnlyHint": True # Only safe operations } ) ``` </Accordion> <Accordion title="Agent Configuration" icon="robot"> Defines agent behavior: ```python Agent( name="Agent Name", model=OpenAIChat(id="gpt-4o"), # LLM model tools=[filtered_tools], # Available tools instructions=[...], # Agent guidance markdown=True # Rich formatting ) ``` </Accordion> <Accordion title="Instructions" icon="list-check"> Guide agent behavior and responses: ```python instructions=[ "You are a [domain] specialist.", "Your capabilities include:", "- Specific capability 1", "- Specific capability 2", "", "Always explain results in business terms.", "Recommend specific actionable steps." ] ``` </Accordion> </AccordionGroup> --- ## Development Workflow ### 1. Design Agent Purpose Define what your agent should do: - What domain does it cover? - What questions should it answer? - What actions can it perform? ### 2. Select Toolsets Choose relevant toolsets for the domain: - `performance` - System monitoring - `sysadmin_discovery` - Service discovery - `sysadmin_browse` - Detailed exploration - `sysadmin_search` - Find specific items ### 3. Configure FilteredMCPTools Set up annotation filters: ```python tools = FilteredMCPTools( annotation_filters={"toolsets": ["your_toolset"]}, debug_filtering=True # See filtering decisions during development ) ``` ### 4. Write Agent Instructions Clear, structured guidance: ```python instructions=[ "You are a [role] for [system].", "", "Your workflow:", "1. First step", "2. Second step", "", "Always [guideline].", "Never [restriction]." ] ``` ### 5. Test and Iterate - Start with debug mode enabled - Test with example queries - Review tool selection decisions - Refine instructions based on behavior --- ## Next Steps <CardGroup cols={2}> <Card title="FilteredMCPTools" icon="filter" href="/agents/agno/filtered-mcp-tools"> Master annotation-based filtering and toolset selection </Card> <Card title="Agent Patterns" icon="diagram-project" href="/agents/agno/agent-patterns"> Learn proven patterns for specialized IBM i agents </Card> <Card title="Advanced Configuration" icon="cog" href="/agents/agno/advanced-config"> Memory, evaluation frameworks, and production deployment </Card> </CardGroup> <Note> **Learning Path**: Read this page for concepts → Follow the [Complete Tutorial](/agents/agno/tutorial) for hands-on setup → Study [Agent Patterns](/agents/agno/agent-patterns) for architecture → Apply [Advanced Configuration](/agents/agno/advanced-config) for production. </Note>