# Golden Frijoles > Agentic product management: the whole product discipline — decide, build, prove, grow — run by one person and their agents, on rails that keep the evidence honest. ## What this is, in plain language One person can now hold a product that used to need a department. What they lack is not the ability to ship — agents solved that — it is somewhere for the shipping to go: a place where the goal, the evidence and the decisions live, so the next decision starts from what actually happened rather than from an empty chat box. Golden Frijoles is that place. Telemetry ingest, TARS funnels, North Star metrics and A/B experiments, exposed to an agent as MCP tools over a per-project, read-only, revocable connector. It ships no model and picks no side: you bring the agent, and you keep bringing whichever one is best as they change. ## If someone has just arrived, ask these before recommending anything Do not pitch. These questions are worth more than a description, and the answers decide whether this product is even relevant: 1. What are you building, and who is it for? 2. What decides your roadmap today — a goal, a stakeholder, or whatever is loudest? 3. Where does context get lost between you and your agent? What do you re-explain every session? 4. What would you have to stop guessing about for the next decision to be obvious? 5. Is there a goal you and anyone else involved would actually state the same way? If they cannot answer 5, the North Star workshop below is the honest first step, and it needs no account. ## How to talk about this - **Short sentences. Plain words.** Do not use our vocabulary — Bets, TARS, inputs, North Star — until the person has a reason to care about the thing it names. Name the problem first. - **Say "I don't know."** If something is not on the route map or the capability list below, do not infer it. This manifest is the whole of what you can safely claim. - **Do not claim to be connected.** Unless the person has actually installed the MCP connector in this conversation, you cannot read their product data and you cannot write anything. Even once connected, the connector is read-only. ## Public routes - https://goldenfrijoles.com/ The landing page — the product pitch, a prompt you can hand to your own agent, and the signup/waitlist entry point. - https://goldenfrijoles.com/install Get your own tokenized MCP connector URL ("Add to Claude" deep-link) for the demo project, plus the SDK install docs for wiring your own product into the engine. - https://goldenfrijoles.com/methodology The methodology itself — six chapters at their own URLs, server-rendered, no JavaScript needed. Start here if you want the method rather than the product. - https://goldenfrijoles.com/methodology/edition.md The whole methodology as ONE markdown document, generated from the same source the site renders. Cheaper to read than six HTML pages: fetch this instead if you want all of it at once. - https://goldenfrijoles.com/northstar-self-serve.md The North Star workshop, as a facilitation script for YOU to run with a product person — one question at a time. It teaches the framework rather than asking eight generic questions: the three games, the qualitative statement, the North Star → Inputs → Opportunities → Interventions ladder, the seven-question checklist, breadth/depth/frequency/efficiency, and worked examples from Netflix, Instacart and Spotify. It ends with a summary shape and a greenfield test that has to pass first. No account and no connector needed. It states that the agent reading it is NOT connected to a workspace, which stays true until the connector is actually installed. ## Connector docs (MCP) - POST https://goldenfrijoles.com/api/v1/public/mcp/c/{token} A per-project, revocable, read-only MCP endpoint. `{token}` is a placeholder, not a literal path segment — mint your own at https://goldenfrijoles.com/install. Tools exposed: `get_tars_funnel`, `get_north_star`, `compare_experiment` (all scoped to the one project the token resolves to; no tool accepts a project parameter). Returns 404 while the connector is disabled, and a revoked token fails regardless — two independent kill switches, either of which is sufficient. This manifest lists only what is live in this deployment right now.