# Optifeed Radar: complete product context Canonical site: https://radar.optifeed.com/ Repository: https://github.com/optifeed/optifeed-radar npm: https://www.npmjs.com/package/optifeed-radar Current verified release: 0.2.3 License: MIT Runtime: Node.js 20 or newer Execution: Agent Skill, local CLI, and local stdio MCP server ## What Optifeed Radar is Optifeed Radar is an open-source AI visibility checker. It asks real AI engines real buyer questions and estimates whether they recommend a brand. It also measures the visibility of products explicitly supplied by a merchant and shows the competing products that appeared for each product's shopper questions. Radar is not a traditional search rank tracker and does not claim a permanent AI ranking. Its results are point-in-time estimates from sampled answers. Answers vary between engines and between runs. The product keeps prompts, captured answers, detected mentions, positions, competitors, and citations available as evidence beneath aggregate results. ## Supported engines - OpenAI / ChatGPT models via `OPENAI_API_KEY` - Anthropic Claude via `ANTHROPIC_API_KEY` - Google Gemini via `GOOGLE_API_KEY` - Perplexity via `PERPLEXITY_API_KEY` The user can configure any one engine or multiple engines. Provider requests use the user's keys. Keys stay in the local environment or local `.env` file and are not sent to Optifeed. ## CLI capabilities `audit `: Free zero-key readiness audit. Checks robots.txt access for AI crawlers, llms.txt, schema.org structured data, metadata basics, and the sitemap. Makes no LLM calls. `check `: Discovers the brand, generates an intent-matched buyer-question pack, asks configured engines, and estimates recommendation, position, and share of voice. Supports `--quick`, `--grounded`, `--json`, `--report`, `--max-cost`, `--fail-under`, and `--yes`. `shopping `: Beta product-level visibility check. The user supplies up to ten products in any order. Radar measures each product against its own category questions, reports visibility and observed shelf position, and lists competitors that appeared instead. Products are not automatically imported or discovered. Input order only breaks ties between identical results; it is not treated as a merchant ranking. Descriptors improve results for opaque model names. `queries `: Show or export the buyer-question pack. `diff `: Compare the two latest saved brand runs without making AI calls. `sources `: Show domains cited in a saved run and the brand's source share without making AI calls. `config`: Report which provider keys are present and where state is stored. It never prints key values. ## MCP tools `audit_store`: Free readiness audit for a domain. No keys and no AI calls. `check_visibility`: Run the brand visibility pipeline. Accepts a domain, optional engines, quick mode, and an optional USD spend cap. The default cap is $0.50. `generate_buyer_queries`: Discover the brand and produce its buyer-question pack for review. Usually requires a small number of model calls. `shopping_check`: Check products explicitly supplied by the user in any order. Accepts optional engines and a spend cap. Results put products that were never recommended first, then sort the remainder by visibility. Beta. `get_snapshot_diff`: Compare the two most recent locally saved brand snapshots. Free. ## Agent Skill and Claude Code plugin Install Radar's workflow in Codex, Claude Code, Cursor, and compatible local coding agents through the Skills CLI: ```bash npx skills add optifeed/optifeed-radar --skill optifeed-radar -g ``` Skills.sh listing: https://skills.sh/optifeed/optifeed-radar/optifeed-radar The skill starts with the free audit, confirms scope and a maximum cost before paid provider calls, and reports evidence and limitations beneath aggregate scores. It uses the CLI when MCP is unavailable and prefers the five structured MCP tools when connected. Installing the skill does not create a public ChatGPT web or mobile app. Claude Code users can install the Agent Skill and local MCP server together: ```text /plugin marketplace add https://github.com/optifeed/optifeed-radar.git /plugin install optifeed-radar@optifeed ``` The Agent Skill is the easiest guided setup. MCP provides structured agent-native tools. The CLI provides direct control for people, scripts, and CI. ## Installation and connection Install or run the public package: ```bash npx optifeed-radar audit yourbrand.com ``` The published package exposes the MCP server as `optifeed-mcp`: ```json { "mcpServers": { "optifeed-radar": { "command": "npx", "args": ["-y", "--package=optifeed-radar@0.2.3", "optifeed-mcp"] } } } ``` Contributors can also run from source: ```bash git clone https://github.com/optifeed/optifeed-radar.git cd optifeed-radar npm install npx tsx src/cli/index.ts audit yourbrand.com npm run build ``` For the cloned MCP server, use `node /absolute/path/to/optifeed-radar/dist/mcp/index.js` in the MCP client configuration. ## Cost and safety The audit and reads of saved results are free. Visibility, query-generation, and shopping operations spend API credit through the user's provider accounts. Costs change with providers, models, prompt count, and grounded web search. Radar reports measured spend and supports explicit caps. A cap is checked before calls; the exact cost of an in-flight call is only known when it returns, so any measured overshoot is disclosed. Start with the free audit. Before an unattended agent starts a paid run, review the generated buyer questions and set a conservative `max_cost`. ## Brand versus product measurement Brand visibility uses unbranded buyer questions to measure whether the brand surfaces without being named. Questions that explicitly name the brand are reported separately as reputation rather than contributing to the main visibility score. Product visibility requires a user-supplied product list. Each product is tested against its own shopper questions, so the result is not a head-to-head ranking between the supplied products. Radar reports a visibility estimate, observed shelf position, and the competing products that appeared. Products that were not recommended lead the summary so the most actionable gaps appear first. ## Grounding and evidence Grounded answers that actually searched the web are reported separately from parametric answers. Requesting a grounded mode does not guarantee an engine will search on every answer. When citations exist, Radar records them alongside the captured answer. ## Stable example The Apple.com example at https://radar.optifeed.com/examples/apple/ is a fixed demonstration dataset used by the landing-page animation. It is explicitly not a live or current claim about Apple. It is available as HTML, Markdown, and JSON so people, crawlers, and agents can inspect the same example in suitable formats. ## Limitations - Point-in-time check, not continuous monitoring. - Scores are estimates and vary between runs. - Product checks cover only products named by the user; there is no catalog import. - Paid checks use the user's own provider credit. - Radar does not guarantee inclusion or ranking in any AI system. - Hosted remote MCP, catalog discovery, and feed-protocol linting are roadmap work, not launch capabilities. ## Documentation - Product: https://radar.optifeed.com/ - Agent Skill HTML: https://radar.optifeed.com/skills/ - Agent Skill Markdown: https://radar.optifeed.com/docs/skills.md - Skills.sh listing: https://skills.sh/optifeed/optifeed-radar/optifeed-radar - MCP discovery: https://radar.optifeed.com/.well-known/mcp.json - Agent skill discovery: https://radar.optifeed.com/.well-known/agent-skills/index.json - Agents: https://radar.optifeed.com/agents/ - MCP HTML: https://radar.optifeed.com/mcp/ - MCP Markdown: https://radar.optifeed.com/docs/mcp.md - CLI HTML: https://radar.optifeed.com/cli/ - CLI Markdown: https://radar.optifeed.com/docs/cli.md - Codex HTML: https://radar.optifeed.com/codex/ - Codex Markdown: https://radar.optifeed.com/docs/codex.md - Claude Code HTML: https://radar.optifeed.com/claude-code/ - Claude Code Markdown: https://radar.optifeed.com/docs/claude-code.md - Methodology: https://radar.optifeed.com/docs/methodology.md - Example HTML: https://radar.optifeed.com/examples/apple/ - Example Markdown: https://radar.optifeed.com/examples/apple/report.md - Example JSON: https://radar.optifeed.com/examples/apple/report.json Last reviewed: 2026-08-03.