Skip to main content
The Gennia CLI (@gennia/cli) is the gennia command you run in your terminal. Three things it solves:
  1. gennia mcp install installs the MCP server into every AI client on your machine (Cursor, Claude Code, Claude Desktop) in one go.
  2. gennia api is the escape hatch into every Public API endpoint — useful for scripts, CI, bash automation.
  3. gennia knowledge upload, gennia skills upload, gennia hub upload-logo cover what the MCP can’t: multipart file uploads.
CLI and MCP complement each other. The MCP is for conversational chat inside your IDE / app; the CLI is for scripts, shell-bound agents, and bulk operations. Same API underneath, same @gennia/sdk internally.

Install

Requires Node 20+.

Authentication

Prompts for your API key (gsk_...) — you paste it, it shows as ******** while typing (raw-mode masking, doesn’t leak into shell history). The credential is saved at ~/.config/gennia/config.json with 0600 perms.
Default = production (api.gennia.ai). If your key was generated in a dev/staging workspace, pass --base-url:
Generate the key inside Studio: open your workspace, Settings → API Keys. To confirm:
Resolution order: --api-key flag → GENNIA_API_KEY env → config file.

Install the MCP into your AI clients

This is the marquee command. It auto-detects which AI clients you have installed (Claude Code, Cursor, Claude Desktop) and writes the MCP server config into the right file for each. It also drops a SKILL.md in Claude Code’s skills directory so the agent “knows” about Gennia in fresh sessions. Non-interactive flags for CI / scripts:
Related commands:

Escape hatch: gennia api

Covers every Public API endpoint. The CLI doesn’t need a custom wrapper for each — gennia api METHOD /path handles all of them.
Paths can be:
  • Relative: /agents → resolves to https://api.gennia.ai/public/api/v1/agents
  • Absolute with prefix: /public/api/v1/agents
  • Full URL: https://api.dev.gennia.ai/public/api/v1/agents

File uploads

Three API endpoints take multipart/form-data instead of JSON. The CLI has dedicated commands:
All support --json and the standard exit codes (see below).

Output contract

Every CLI command follows the same rules so it’s reliable in scripts and safe for AI agents:
  • stdout = data. Stable JSON by default in non-TTY, pretty in TTY. Force JSON with --json.
  • stderr = humans. Banners, progress, hints. Suppress with --quiet.
  • NO_COLOR=1 disables ANSI colors on any command.
  • Exit codes that mean something:
  • Errors are always structured. In --json mode:
  • No interactive hang. Every command that needs input detects non-TTY and fails fast with a hint pointing at the missing flag.

Environment

Bundled skill file

Every CLI install ships a skill.md (~6 KB) that gennia mcp install copies into ~/.claude/skills/gennia/SKILL.md (and the Cursor equivalent). It’s a markdown file that teaches the AI agent:
  • When to reach for MCP tools vs gennia api in the shell
  • Domain → endpoint mapping
  • Common workflows (list every agent, bulk update, etc.)
  • Caveats (uploads need file_path, destructive mutations ask for confirmation)
Net effect: the first session of any AI client on your machine already “knows” about Gennia, without you having to explain.

Update

Or see GitHub releases. Packages are signed with SLSA provenance (the “Built and signed on GitHub Actions” badge on npm).

Source

github.com/Gennia-Tech-Group/gennia-cli — MIT, public.

Next steps

Install the MCP

The CLI already installs the MCP via gennia mcp install, but it’s worth reading how the MCP server works and what you gain from it.

Full reference

Every endpoint that gennia api calls is documented here with schemas and try-it-out.