SDK¶

Platforms: claude openai gemini m365-copilot

What an SDK Is¶

An SDK (Software Development Kit) is a framework or toolkit that provides abstractions for building AI workflows in code. Where APIs give you raw access to models and services, SDKs give you patterns and structure — handling orchestration logic, tool use, memory management, and multi-agent coordination so you don't wire everything from scratch.

Think of it this way: an API lets you call a model. An SDK lets you build an agent that calls models, uses tools, manages state, and coordinates with other agents — with the plumbing handled for you.

Key Characteristics¶

• Higher-level abstractions — wrap raw API calls with patterns for tool use, memory, multi-turn conversations, and agent loops
• Handle orchestration logic — manage the planning, routing, and error recovery that make agents work reliably
• Opinionated about patterns — provide built-in support for agent loops, handoffs between agents, guardrails, and structured outputs
• Multi-language support — most SDKs offer Python as a primary language, with TypeScript/JavaScript as a secondary option

When to Use It¶

Use SDKs when:

• You're building agents or multi-step workflows in code and want established patterns rather than building from scratch
• You need tool use orchestration — the SDK handles the loop of "model decides to call a tool → tool executes → result goes back to model"
• You're coordinating multiple agents that need to hand off work to each other
• You want guardrails, memory management, or structured output validation built into the framework
• You're moving beyond a single API call into workflows that require state, iteration, and decision-making

Example¶

Using the Claude Agent SDK to build a research agent that takes a topic, searches the web, reads relevant pages, evaluates source quality, and produces a structured report — the SDK handles the agent loop (plan → act → observe → repeat), tool execution, and conversation state, while you define the tools and instructions.

Or using LangGraph to orchestrate a content pipeline where a research agent hands off findings to a writing agent, which hands off a draft to an editing agent — each with different tools and instructions, coordinated through a graph-based workflow.

Frameworks¶

Agent-to-Agent (A2A) Protocol¶

The Agent-to-Agent (A2A) protocol is an open standard for agent interoperability. It defines how agents built with different frameworks can discover each other, negotiate capabilities, and collaborate on tasks — regardless of which SDK or platform they were built on.

A2A complements MCP: where MCP connects agents to tools and data, A2A connects agents to other agents.

A2A specification and documentation

Key Concepts¶

Agent loop — The core pattern in most SDKs: the model receives a goal, decides what to do, calls tools, observes results, and repeats until the goal is met or a stop condition is reached.

Tool definition — SDKs provide structured ways to define tools the agent can call — functions, APIs, file operations, web search — with schemas that tell the model what each tool does and what parameters it accepts.

Handoffs — In multi-agent workflows, one agent passes control (and context) to another. SDKs formalize this with handoff protocols that manage what information transfers between agents.

Guardrails — Built-in safety checks that validate inputs and outputs at each step. SDKs let you define rules (content filters, output validators, scope limiters) that run automatically during agent execution.

Memory and state — SDKs manage conversation history, tool call results, and workflow state so the agent can reference prior steps. Some support persistent memory across sessions.

Structured outputs — Many SDKs integrate schema validation so agent responses conform to expected formats — critical for pipelines where one agent's output feeds into the next.

Platform Implementations¶

Relationship to Other Blocks¶

SDK is the code-first orchestration layer:

• Model is what SDKs orchestrate — they manage the model's decision-making loop
• API is what SDKs call under the hood — they abstract raw API calls into higher-level patterns
• Prompts are embedded in SDK agent definitions as system instructions and tool descriptions
• Context is managed by the SDK through memory, conversation history, and tool results
• Skills are conceptually similar to SDK tool definitions — both package capabilities for the agent to invoke
• Agents are what SDKs build — the agent concept comes to life through SDK code
• MCP servers are a common tool type in SDKs — agents use MCP to connect to external systems

Related¶

• Agentic Building Blocks — SDK in the context of all nine building blocks
• API — the raw programmatic interfaces that SDKs abstract over
• Agents — the building block that SDKs implement in code
• Agent Capability Patterns — architectural patterns that SDKs help implement
• Multi-Agent Collaboration — patterns for coordinating multiple agents
• MCP — the protocol that connects SDK-built agents to tools and data
• AI Use Cases — what teams build with these blocks
• Coding Use Cases — code-first AI workflows
• Workflow Architecture Patterns — architectural patterns from augmented LLMs to autonomous agents
• Platforms — platform-specific SDK guides