Official scripts
Explore our curated collection of scripts
Delete Perplexity chat
Permanently delete one Perplexity thread from your recent history, by the chatId returned from list_chats.
Get Perplexity chat messages
Full message transcript of one Perplexity thread, by the chatId returned from list_chats.
List Perplexity chats
List your recent Perplexity threads (most recent first), as shown in the account's sidebar.
Search Perplexity chats
Search your Perplexity thread history by keyword, via the Sessions library's search box.
Ask Perplexity
Ask Perplexity a question and get back three things the answer alone cannot tell you apart: the answer itself, the queries Perplexity's own search issued for it, and every page that search returned — not only the ones the answer cited. That split is the point. `sources` is the candidate pool, in the order the thread's Links view lists it, each with its publisher, title and snippet. `citations` is what the answer actually cited inline, each with the real destination URL, the chip as shown (a "+N" groups further sources that are not exposed individually) and whether Perplexity marks the domain as trusted. So `sources` minus `citations` is what Perplexity retrieved and then passed over, while a URL in neither was never retrieved at all — which is what you need to see how a topic, brand or product gets described and who gets credited. Headings, bullet lists and comparison tables come back as readable text, and `threadUrl` links the thread so any answer can be audited afterwards. Treat one run as a single draw rather than a measurement: the same question can search differently, answer differently and cite different pages next time, so ask more than once before concluding anything. Perplexity's own free-tier ask rate limit can gate a request instead of answering it, answering "Sign up and repeat your request." instead. This can happen whether or not the browser is signed in — a live test confirmed the gate firing on a genuinely signed-in free-tier account, so signing in is not a guaranteed fix, only a paid plan is known to avoid it. The script names that state rather than returning an empty answer.
Ask Claude
Open a new Claude.ai chat, send one question, and wait for the answer to finish (up to 3 minutes). Returns the answer as the markdown the model emitted, the message and conversation ids, why generation stopped, which model answered, the web_search queries it issued, every source its web_search and web_fetch tools surfaced, and — separately — the sources the answer actually cited. Requires a signed-in session: claude.ai has no anonymous chat. Avoid running many of these at once: claude.ai limits how many chats one account may hold concurrently and refuses the extras, so a refused send comes back quickly carrying Claude's own message and only needs to be retried at a slower pace.
Get Claude chat messages
Full message transcript of one Claude.ai conversation by uuid, oldest→newest. `text` is the readable prose; `tools` lists tool calls the UI renders as cards.