tool use patterns

MCP is the protocol layer enabling Claude to use tools. It standardizes how tools are defined and called, making tool use systematic rather than ad-hoc.

MCP is the infrastructure layer that enables standardized tool use. The article explains how MCP enables LLMs to access external services/tools in a consistent way.

Demonstrates how tool design should be shaped by context constraints, not just capability. Tool design is context engineering.

Agent skills are implemented as tools. The pattern suggests tool selection should be curated to problem clarity, not enumerated exhaustively.

The tool exposes web operations (search, fetch, image) as structured LLM tools, demonstrating how practitioners design tool interfaces for context augmentation.

Agent architectures critically depend on how tools are invoked, how outputs are captured, and how that context re-enters the reasoning loop—core tool-use pattern analysis.

Dedicated section on 'Creating Tools' and 'Tools Calling' demonstrates how context flows when agents delegate to external functions.

MCP is the standardized infrastructure layer that defines how tools are exposed to and discovered by AI agents

MCP server is a tool interface. Author is using tool-use pattern to extend Claude's capabilities into fitness app domain.

MCP standardizes how AI systems discover and use external tools, making tool-use a persistent capability rather than re-explained each turn

Content explicitly discusses basic tool use as starting point, describing how agents invoke and integrate tool results

Explicitly discusses tool orchestration as first-class concept with workflow patterns (CRM checks, email drafts, ticket updates).

Extends tool-use patterns by showing how to decouple tool output from context inclusion

Tool definition design (Layer 4) is presented as critical to agent decision-making; overlapping tool definitions waste tokens. This directly supports tool-use effectiveness.

MCP serves tools to agents on demand; the optimization strategies (tool search without consuming token window) represent a new pattern for intelligent tool discovery

Standard tool use in agents requires tool definitions in context. MCP-based execution extends this by outsourcing the execution layer, changing the context economics.

Demonstrates tool integration via connector pattern for enabling Claude action on external services

Agent tool composition is a specific instantiation of how agents use external tools to overcome capability gaps

MCP enables systematic tool integration through protocol, relevant to how AI agents access external capabilities

MCP servers enable standardized tool integration, making tools discoverable and usable across agents via common context protocol.

repo_updater tool is a mechanism for centralizing specialized capability and preventing agents from doing it poorly

MCP is fundamentally about standardizing how AI systems access and use external tools

Article shows how developers integrate external tools (GitHub, databases, browsers) as context sources for AI assistants

All frameworks discussed include 'tool use and calling' as a key component, confirming this as a fundamental capability in agent systems.

Section on 'Creating Tools' shows how tools constrain agent capabilities and define agent-to-tool context boundaries. Tools filter what information agents can act upon.

Agent architecture relies on tool calling as the primary action mechanism. Context engineering must optimize for tool description, selection, and result integration.

Tutorial centers on agents calling tools and other agents; demonstrates tool integration as context distribution mechanism

MCP servers enable tool use at the protocol level, allowing structured interaction with external systems through the conversation interface.

Shell and GitHub access as controlled tool interfaces reveals how to structure tools within constrained context for AI safety and efficiency.

ReAct pattern and tool calling explicitly mentioned as agent capability layer

Addresses how tool calling works across different agentic frameworks, which is a foundational pattern in agent architecture

Research assistant using Firecrawl and coding tools implies patterns of tool sequencing and state passing between tool calls.

Shows how tool protocols (MCP) affect human-AI collaboration patterns; UI support changes how users interact with external systems via AI

Shows how Claude Code uses MCP tools (feature flag APIs) within a structured integration

Agents primarily work by using tools; the content teaches how LangChain implements tool calling, which is fundamental to how context flows through agent systems.

Describes tool wrapping pattern (Playwright-style page/locator/browser objects as JS functions) that agents call

Claude agents reliably implementing arbitrary libraries is a specific tool-use pattern. Suggests repeatable context strategy for library/integration generation.

Slash commands (/spec, /plan, /build, /test, /review, /ship) are tool invocations that trigger context-specific skill activation

Each framework handles tool calling and result integration differently, which affects how context is preserved after tool execution

Multi-agent systems often coordinate through tool use; orchestration patterns affect how tools are routed between agents and how results are contextualized.

Framework discussion includes how agents access tools, which relates to information availability patterns

Agents use tools as part of reasoning loops; tool context and integration affects information flow

Article mentions agents can have tools; tool routing is part of context management in multi-agent systems

Agent autonomously selects which database collections to query, suggesting tool-use planning, but implementation details are hidden.

Lists tool/action orchestration as common feature but provides no guidance on implementing or optimizing tool integration

Agents using email API as tool to interact with humans, but no technical breakdown of tool integration or context management