Building tools

For everyone who builds tools with an AI agent: access tokens, workspace, draft and publishing.

• OverviewThe way from an idea to a released tool, in seven steps.
• Access tokensAn access token signs in an agent such as Claude Code under your name.
• WorkspaceOne command clones the tool repository and registers your instance with the agent.
• Working with the agentWhat an agent can do in your instance over MCP, what it can never do and what it needs to know from you.
• Collections and fieldsField types, technical columns and the rule that a collection only grows.
• Layout of a toolWhich files a tool has, what tool.json says and what flow.js brings.
• Draft and publishingEverything that changes a tool goes to the draft. A person puts it live.
• Developing locallyBuild and view a tool on your own machine with new-tool.py and dev-proxy.py.
• Reading and writing dataThe calls of the data interface: query, page, write, evaluate, export.
• Files, email and peopleStore files in buckets, send email, look people up and read your own permissions.