How matlab/matlab-ai-agent-sdk works
MATLAB AI Agent SDK lets MATLAB users call language models directly or create stateful agents that can use tools and agent-workspace data. The principal workflow is to create a client, create an agent with a prompt and optional tools, then run it to obtain a response and retain conversation history. The tree also shows distinct skill and MCP capabilities. Only README-described relationships are evidenced; source excerpts were unavailable, so uncertain internal calls are omitted or marked dashed.
Components 20
Agent Runtime
- AI Agent
- Message Types
Model Clients
- Client Factory
- Client Base
- OpenAI Chat
- OpenAI Responses
- Ollama Client
Tools and Skills
- Tool Definition
- Tool Arguments
- Local Tool
- Built-in Tools
- Tool Approval
- Load Skill Tool
- Skill Registry
MCP Integration
- MCP Client
- MCP Tool
Other components
- MATLAB User
- OpenAI API
- Ollama
- MCP Server
Connections 22
- MATLAB User → Client Factory: creates client
README.md - MATLAB User → AI Agent: creates and runs
README.md - Client Factory → OpenAI Chat: selects API
- Client Factory → OpenAI Responses: selects API
- Client Factory → Ollama Client: selects API
- AI Agent → Client Base: uses client
- OpenAI Chat → OpenAI API: requests completions
README.md - OpenAI Responses → OpenAI API: requests responses
README.md - Ollama Client → Ollama: connects to models
README.md - AI Agent → Message Types: tracks history
README.md - MATLAB User → Tool Definition: defines tool
README.md - Tool Definition → Tool Arguments: describes arguments
README.md - AI Agent → Tool Definition: uses tools
README.md - AI Agent → Local Tool: runs tool
- AI Agent → Built-in Tools: uses tools
- AI Agent → Tool Approval: requests approval
- MATLAB User → Load Skill Tool: loads skill
- Load Skill Tool → Skill Registry: registers skill
- MATLAB User → MCP Client: uses MCP client
- AI Agent → MCP Tool: uses MCP tool
- MCP Tool → MCP Client: delegates to
- MCP Client → MCP Server: communicates with