A single pass at a strategy question gives you the obvious answer. Running it over several cycles with contradictions surfaced between them is a different exercise, and here you set how many — one for a quick read, five when the question deserves it. The continuation tool exists because these answers get long, which is a fair signal about what you are asking for. No key at all is the other reason to keep it around: nothing to provision before you try it.
An MCP server that routes two kinds of work to the Pollinations AI API: extended reasoning on a question you pose, run over several cycles with contradiction detection between them, and real-time web search through a search-capable model. Both go through public endpoints, so there is nothing to sign up for.
- A multi-cycle analysis of a topic, question or problem, with the number of cycles under your control — default 3, maximum 5 — and the model selectable; a `seed` makes a run reproducible — `think`
- Live web search returning current results with their sources, for the questions where anything older than today is wrong — `search`
- Long answers handled rather than truncated: responses past roughly 30KB are split, and the next part is fetched on request while the continuation state is held across calls — `continue_thinking`
Nothing — no account, no API key, no environment variables; it uses public Pollinations.ai endpoints. Clone the repository, run `npm install`, and point your client at `index.js` by absolute path. Since the endpoints are public and shared, treat availability as best-effort: the server has retry logic and falls back rather than assuming every call succeeds.