Perception tracks 590+ Bitcoin and crypto accounts on X: founders, developers, fund managers, analysts, journalists, and the companies themselves. The corpus holds 660,000+ posts with coverage back to 2016, and roughly 50,000+ new posts land every month as tracked accounts publish.

Every post carries an AI sentiment label (positive, neutral, or negative), detected entities with their own per-entity sentiment, the canonical author handle, and timestamps. You query by keyword, account, date range, or sentiment and get structured JSON back. Raw text collection is the easy half of the job; the labels are what make the data usable in analysis.

The X corpus sits inside a larger index: 1.8M+ items across 1,000+ curated sources including news outlets, podcasts, conference talks, regulatory filings, and GitHub activity. That means you can compare what crypto Twitter is saying against what the press and the podcast circuit are saying, in the same query language. The sentiment API and the LinkedIn intelligence corpus use the same key.

The official X API works, and at historical scale it is priced for enterprises. Getting years of posts from hundreds of accounts through the paid tiers is a real line item, and firehose-level access costs more than most research budgets.

Scraping is the route people reach for next, and it carries costs that only show up later. Scraping violates X's terms of service. Accounts get banned, IPs get blocked, and the platforms actively litigate against scrapers. The engineering side is its own treadmill: proxies, parsers, and an anti-bot arms race that breaks your pipeline every few weeks. You end up maintaining infrastructure instead of doing analysis.

And after all that, raw tweets are unlabeled text. Sentiment scoring, entity extraction, deduplication, author normalization: that enrichment layer is a second project you still have to build and pay for.

Our position is simple. If a collection method feels like it might get you in trouble, it will, eventually. Pay for the data instead.

Generate a key from Settings, then authenticate with a Bearer token:

curl "https://api.perception.to/feed?keyword=bitcoin" \
  -H "Authorization: Bearer pcp_your_key_here"

Filter by outlet to isolate X posts, or by sentiment to pull only negative or positive coverage. Responses include sentiment breakdowns and per-outlet tone, so one call answers both "what was said" and "how it landed."

Or skip the code entirely. The same key connects Perception to Claude, ChatGPT, and Gemini over MCP, where the search and sentiment tools become native functions your AI assistant calls mid-conversation. Full endpoint reference in the documentation, and the broader data surface is covered on the data API page.

Voice dashboards. Track what specific founders, analysts, and funds are posting, with per-voice sentiment timelines, so you see who turned bearish and when.

Research notebooks. Pull historical posts by keyword and date range into pandas and measure how the tone of crypto Twitter moved around past market events. Measurement of what happened, with labeled data instead of raw text.

Comms monitoring. Watch mentions of your company or competitors across tracked accounts, and catch a shift in tone before it reaches the press.

AI agents. Connect over MCP and ask in plain English: "what did tracked accounts say about ETF flows this week, and what was the sentiment split?" The agent runs the queries itself.