Skip to main content

Agent Tooling

How do you build a tool an agent can operate without guessing?

This is the engineering side of the toolkit — for the builder who has to transform a capability into something an agent can call reliably. If instead you want to choose which tools bring value to a workflow, start with the AI Toolkit (the business-value view). This hub is the how-to-build view.

The distinction that governs everything here: the agent is the first customer. A CLI or server that a human can muddle through is not one an agent can. Agents cannot read prose output, cannot infer undocumented flags, and hallucinate differently than humans typo. Agent-grade means predictable contracts, runtime introspection, and safe rehearsal — designed in, not bolted on.

The Spine​

  1. Agent-First CLI Design — turn a human-first CLI into one an agent can drive: JSON contracts, input hardening, dry-run safety, and the retrofit order that minimises risk.
  2. Agentic CLI Checklist — the twelve-dimension audit that stress-tests whether a CLI is truly agent-grade, with a scoring sheet and release decision.
  3. MCP Server Engineering — the architecture, three primitives, transports, and token economics of building a Model Context Protocol server that any client can reuse.

The Two Jobs​

  • Appreciate value — which tool, what it does, whether to adopt it. That is a selection job for a decision-maker → AI Toolkit.
  • Transform value — how to build the tool, the contract it exposes, the cost it imposes. That is a build job for an engineer → this hub.

The same tool appears in both views. The AI Toolkit tells you Firecrawl is worth adopting for web extraction; this hub tells you how to expose your own capability so an agent adopts it the same way.

Context​

  • instance-of Intelligent Hyperlinks — a tool contract is the third pipe: a protocol that lets an agent discover, call, and verify a capability
  • pairs-with AI Harness — the harness is the agent side of the contract; agent tooling is the tool side
  • pairs-with Agent Protocols — MCP, A2A, and the broader standard stack these tools speak
  • applies-to AI Toolkit — the business-value catalogue this engineering serves
  • up Software — the engineering knowledge base this hub belongs to

Zoom Out​

  • Up: Software — the full engineering wiki
  • Across: AI Toolkit — choose the tools; this hub builds them