Skip to main content

AI Toolkit

Which tool would remove friction from this workflow without hiding human judgment?

The toolkit is where you choose the practical aids for directing attention: which model to use, which app to work in, which prompt or skill to reuse, and which tool connection should enter the loop. This is the value-appreciation view — what each tool does and whether it's worth adopting. For the engineering view — how to build an agent-grade tool — see Agent Tooling.

Who This Is For​

Use this section after you know the workflow you are improving. If you do not know the workflow yet, start with Work Mapping.

Where To Start​

Pick by the job you're trying to do:

  • Compare model providers and modalities → Models
  • Work through a conversational interface → Chat
  • Shape instructions and reusable prompt patterns → Prompts
  • Package repeatable procedures for agents → Skills
  • Connect agents to external systems → MCP Servers

Current Tools​

The CLI surfaces currently relevant to our stack, or adjacent enough to evaluate now:

  • Firecrawl CLI (web data) — scrape, crawl, search, and browser automation from the terminal. High relevance — live web extraction, competitor analysis, research.
  • Google Workspace CLI (workspace ops) — Sheets, Drive, and Gmail automation from the terminal. Medium — when workflows move out of the UI.
  • GitHub CLI (code + repos) — issues, PRs, releases, repo operations. High — a lean default that overlaps with GitHub MCP.
  • Gemini CLI (coding agent) — terminal-native coding with Gemini models. Medium — a comparison surface for large-context or low-cost coding.
  • llm (evaluation + logging) — run prompts, chain outputs, log every run to SQLite, query via Datasette. High — fills the prompt-logging and evaluation gap.

llm is the most recent addition — it closes the evaluation gap: every prompt run is logged and queryable, a direct complement to measuring agent output quality.

Connect Agents to External Systems​

MCP is how agents reach databases, APIs, and services. For choosing servers, start at the MCP Servers hub — the adoption radar, team profiles, and server catalogue. For building a server, see MCP Server Engineering.

  • AI Agents — when a tool becomes a coordinated agent crew.
  • Ship It — when toolkit choices need evals, observability, and config.
  • Agent Tooling — when you need to build the tool, not just choose it.

Context​

  • pairs-with Agent Tooling — the engineering view: how to build the CLIs and servers this page helps you choose
  • instance-of Intelligent Hyperlinks — every tool connection is a pipe that carries a capability into the loop
  • depends-on Models — the intelligence the tools direct
  • applies-to Work Mapping — the workflow a tool choice is meant to improve
  • pairs-with Flow Engineering — how tools serve the development loop
  • up Applications — the applications knowledge base

Questions​

Which tool choice would remove friction from the workflow without hiding human judgment?

  • If the primary operator is already an agent, which of these tools is easiest for it to adopt without help?
  • Which tool are you carrying out of habit that no current workflow actually demands?
  • Where would one well-chosen MCP server replace three brittle custom integrations?