Optifeed radar

Optifeed Radar

Optifeed Radar is an open-source CLI and MCP server that asks real AI engines real buyer questions and measures whether a brand gets recommended. It runs locally, uses the user's own API keys, and stores nothing on a server.

Requirements

• Node >= 20.
• For the check pipeline, at least one engine API key in the environment: OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY (Gemini), or PERPLEXITY_API_KEY. The zero-key audit needs no keys.

Commands (CLI)

Run it with npx optifeed-radar <command> (no clone needed). From a clone, use npx tsx src/cli/index.ts <command> instead.

• audit <domain> - zero-key AI-readiness check (robots.txt, llms.txt, schema.org, meta, sitemap). No AI calls, costs nothing.
• check <domain> - the full pipeline: generate buyer prompts, ask the engines, score recommendation, position, and share of voice into one AI Visibility Score. Needs at least one API key.
• shopping <domain> --products "A, B, C" - product-level check (beta) for the products the user NAMES, in any order. Each product gets its own 0-100 visibility score and the report is sorted by what the engines did, with any product they never recommended first. --products-file products.yml takes a name, aliases, and a descriptor per product; the descriptor ("quiet home espresso machine") is what rescues an opaque product name. Max 10 products per run. There is no catalog import, so ask the user which products to check.
• diff <domain> - what changed between the last two saved runs.
• sources <domain> - domains the AI cited, and the brand's share of voice.
• queries <domain> - show or export the buyer-prompt pack.
• config - which engine keys are set and where state is stored (never the key value).

Useful check flags: --json, --report report.html, --max-cost 0.50, --quick, --grounded, --fail-under 50, --yes (skip the cost prompt so an AI agent can run it unattended).

Tools (MCP)

The optifeed-mcp server exposes the same capability to AI agents:

• check_visibility - run a visibility check for a domain.
• audit_store - run the zero-key readiness audit.
• generate_buyer_queries - produce the buyer-prompt pack.
• shopping_check - product-level check for products the user names, in any order.
• get_snapshot_diff - compare two saved runs.

Worked example

npx optifeed-radar audit example.com
export OPENAI_API_KEY=...   # or put keys in a .env file in this directory
npx optifeed-radar check example.com --quick --yes

The CLI loads .env from the directory it runs in; an exported key wins over the same key in .env. config reports which keys were found and which file they came from, never the values.

How long to expect (measured 2026-07-22): audit takes about a second, check --quick across four engines takes 47 to 51 seconds, and about 97 seconds with --grounded. That wait is the engines answering, not a hang; check streams per-phase progress to stderr while it runs.

Honesty and cost

BYO keys. Scores are estimates from sampling and say so; engines vary between runs. Grounded engines (which cite sources) are reported separately from parametric ones (which answer from model weights alone), and an engine counts as grounded only for the answers where it actually searched - asking for grounded mode is not the same as searching, and the report says so when an engine searched on only some of its answers. Keys stay on the user's machine and are never logged or stored.

Cost, measured on real runs (2026-07-20, --quick = 8 buyer prompts): the zero-key audit is free; a single-engine check is about $0.09; all four engines is $0.41 to $0.46; adding --grounded takes that to about $1.09, since web search is billed on top of tokens. Every run reports what it spent, split into setup and engine calls. --max-cost caps spend and is checked before every call, but a call's cost is not known until it returns, so a run can exceed the cap by at most one unmeasured call per engine; any overshoot is always reported. Confirm the cost with the user before running check on a large prompt pack or with --grounded.

A shopping run is bigger than a check: up to 4 prompts per product on every engine with a key (products that share a category share their category questions, which are asked once), so start with two or three products and --max-cost. No shopping run has been measured across multiple engines yet, so do not quote a cost figure for it. What IS fixed: the MCP tool caps at $0.20 per product by default.

Catalog discovery (importing products from a store or a feed) and product-feed linting are on the roadmap, not shipped - join the waitlist at optifeed.com.

More at optifeed.com