fluxra-cli

Hook generation, AI research, and comment scraping for the terminal. Built for AI agents (Claude Code, Cursor, Codex, OpenClaw) and humans.

Install

npm install -g fluxra-cli
fluxra login --token flx_xxxxxxxxxxxxxxxxxxxxxxxx

Get the token from https://fluxra.app/dashboard/api-keys.

MCP server (Claude Desktop / Code / Cursor)

fluxra mcp runs the same surfaces as a Model Context Protocol stdio server. The host model calls them as tools — no copy-paste, no shell command memorisation.

Claude Desktop — ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "fluxra": {
      "command": "npx",
      "args": ["-y", "fluxra-cli", "mcp"],
      "env": { "FLUXRA_TOKEN": "flx_xxxxxxxxxxxxxxxxxxxxxxxx" }
    }
  }
}

Claude Code:

claude mcp add fluxra -- npx -y fluxra-cli mcp

Cursor: same JSON shape in ~/.cursor/mcp.json.

Tools exposed: fluxra_scrape, fluxra_search, fluxra_hooks_generate, fluxra_ask, fluxra_scrapes_list, fluxra_scrapes_get, fluxra_insights, fluxra_platforms, fluxra_whoami. Run fluxra mcp --help for descriptions.

Quick examples

# Scrape a thread from any supported platform — saves to your
# dashboard (https://fluxra.app/dashboard/scrapes/<id>).
fluxra scrape "https://reddit.com/r/SaaS/comments/abc/foo"
fluxra scrape "https://news.ycombinator.com/item?id=12345"

# Full research pipeline: Reddit search + scrape every result + save.
fluxra research "B2B SaaS pricing pain" --limit 10

# Browse what's already in your dashboard.
fluxra scrapes list --platform reddit
fluxra scrapes get <id>

# Generate 5 viral hooks from a topic prompt (Reddit-grounded).
fluxra hooks "B2B SaaS founders losing trial users in week 1"

# Pro-only AI Ask with web-grounded citations.
fluxra ask "What pricing model do top self-serve SaaS use in 2026?"

# Raw Reddit search via Exa, no AI rewriting.
fluxra search "crm software frustrations" --limit 12

# Account info.
fluxra whoami
fluxra keys list
fluxra platforms

See fluxra --help for the full surface, or SKILL.md for agent-targeted usage notes (Claude Code / Cursor / Codex / OpenClaw all auto-discover the file).

Plan caps

Each batch returns 5 hooks by default (server-side env tunable).

Authentication

The CLI uses per-device API keys (not your account password):

1. Sign in to https://fluxra.app/dashboard/api-keys
2. Click Create key, name it ("My laptop", "CI server"), pick scopes
3. Copy the once-shown raw token (flx_…)
4. fluxra login --token <paste>

Saved at ~/.fluxra/config.json (chmod 0600). Revoke from the same dashboard page when a device is lost — fluxra logout only removes the local copy, the key keeps working on other devices unless revoked server-side.

The dashboard also shows per-key audit logs (action, count, time, IP fingerprint) so you can spot a key being used from somewhere it shouldn't.

Quoting URLs (zsh users)

zsh treats ? as a glob, so unquoted URLs with query strings break with zsh: no matches found. Always quote:

fluxra scrape "https://news.ycombinator.com/item?id=12345"

Or alias once:

echo "alias fluxra='noglob fluxra'" >> ~/.zshrc && source ~/.zshrc

bash users don't have this issue.

Output

JSON by default. --format csv and --format text for the formats that make sense for the command. --output <file> writes to disk instead of stdout, with a one-liner on stderr confirming the byte count — safe to pipe.

API endpoint

All commands talk to https://api.fluxra.app — no configuration needed. The CLI is fully managed; there's no self-host path.

Licence

Apache-2.0.