Use the CLI with AI agents

Give Claude Code, Cursor, or any terminal-capable AI agent safe, scoped access to your Zanfia workspace.

4 min readLast updated Jul 9, 2026

The CLI is designed to be driven by AI agents as much as by humans: every command has a --json mode, errors carry machine-readable reasons, and authentication works from an environment variable. An agent that can run terminal commands can run your whole workspace.

Prefer MCP when your client supports it

If your AI client speaks MCP (Claude, Cursor, VS Code, ChatGPT), the MCP connection is the more direct route — typed tools, no shell required. The CLI approach below works everywhere else, and is the right choice for scripted pipelines and CI.

Setup

Create a scoped API key

In the dashboard, Integrations → API, MCP and CLI → create a key with only the permissions the agent needs (see Authentication). For a content agent that's typically Products read/write; add Communities only if it should post.

Install the CLI where the agent runs

npm install -g @zanfia-dev/cli
export ZANFIA_API_KEY=YOUR_SCOPED_KEY

Tell the agent how to use it

Paste instructions like the block below into your agent's configuration (a CLAUDE.md, Cursor rules, a system prompt, …).

Instruction block to give your agent

## Zanfia workspace access

You can operate my Zanfia workspace with the `zanfia` CLI (beta binary: `zanfia-dev`).
Authentication is already configured via ZANFIA_API_KEY.

- Always pass `--json` (before the subcommand) and parse the output.
- Discover ids first: `products list`, `offers list`, `communities list`.
- Course lessons: `courses get <productId>` for structure; lesson content is
  Markdown; use `--content-file` for long bodies; new content starts as draft —
  publish explicitly with `courses publish`.
- Money is in minor units (1999 = 19.99).
- Destructive commands (`courses lesson delete`, `courses module delete`,
  `pages delete`, `clients revoke`) are permanent — confirm with me first.
- If a command fails with `insufficient-scope`, the API key lacks that
  permission — tell me instead of retrying.

Reference (fetch when unsure):
- Docs index for agents: https://zanfia.com/llms.txt
- Any docs page as raw Markdown: append `.md` to its URL, e.g.
  https://zanfia.com/help/developer-cli/commands.md
- Workspace API OpenAPI spec (all routes + schemas):
  https://zanfia.com/openapi/workspace-api.json

Adjust the confirmation rules to your risk tolerance — the key's scopes are the hard boundary; the instructions are the soft one.

Patterns that work well

Content pipelines — agent drafts lessons in Markdown files, then courses lesson update … --content-file and a final human courses publish.
Community digests — communities posts --json on a schedule, summarize, post the digest with communities post create.
Sales reporting — stats overview --period 7d --json piped into whatever writes your Monday report.
Bulk operations — anything repetitive over ids from a list command; the strict-permutation reorder commands make ordering changes safe to script.

Was this article helpful?

Spotted something off? Tell us at support@zanfia.com.

Use the CLI with AI agents

Setup

Create a scoped API key

Install the CLI where the agent runs

Tell the agent how to use it

Instruction block to give your agent

Patterns that work well

Related articles

CLI quickstart

Connect AI clients to your workspace

Authentication and API keys