Skip to content
AX Principles
AX-SYS-10IntermediateDraft

Expose only the tools the agent needs

Expón solo las herramientas que el agente necesita

#saas#mcp

Scope tools by domain and give each one a clear, single purpose with structured results. Do not overwhelm an agent with hundreds of operations.

Why

Scoping tools by domain improves reliability; dumping 300 operations on one agent degrades it (Speakeasy Toolability). Anthropic's production guidance is explicit: build a few thoughtful tools targeting specific high-impact workflows. Tools occupy space in the context window — every tool listed is a choice the model must evaluate, and a cluttered toolset degrades selection accuracy and increases the chance of misdirected calls.

Do

Provide domain-scoped toolsets.

Don't

Ship one mega-toolset for every agent.

Artifact

Config artifact: expose domain-scoped toolsets (for example orders, billing) rather than one toolset of 300 operations. Gate which set an agent receives by its scopes.

Citations

  1. [01]
    Speakeasy — Designing Agent Experience

    Toolability: scoping tools by domain and giving each a clear single purpose improves agent reliability; exposing too many operations degrades selection accuracy

  2. [02]
    Anthropic — Building Effective AI Agents

    Build a few thoughtful tools targeting specific high-impact workflows; a large, poorly-scoped toolset degrades agent decision quality and increases latency

  3. [03]
    Anthropic — Writing Effective Tools for AI Agents

    Tools are prominent in Claude's context window; developers should be conscious about how they design tools to maximize context efficiency and minimize misdirected calls