Changelog

The MCP server we launched in beta last week is now switched on for every account, on every plan - including the free one. Connect Claude, Cursor, Codex, ChatGPT or any other MCP client and ask what's worth replying to; the setup guide for every client is at /docs/mcp. Plans shape the depth, not the door: everyone gets their matches, tags and counts; Patrol and above lift the result caps, add time filtering, include your drafted replies and add static API keys for headless and scheduled clients. As always, it's read-only - your AI can see your triage inbox, and it can't change anything or post on your behalf.

You can now connect your AI to ThreadScope. Our new MCP server gives Claude, Cursor, Codex, ChatGPT - anything that speaks MCP - read-only access to your triage inbox, so you can ask "what's new and worth replying to?" and get a brief instead of opening a dashboard. Your AI sees your matches, tags, statuses and drafted replies, with links back to Reddit for the threads themselves; it can't change anything or post on your behalf, and you can revoke access any time from Settings, then Connections. Chat clients sign in with OAuth; config-file clients use an API key. The full setup guide for every client lives at /docs/mcp. We're rolling access out gradually in beta - reply to any of our emails if you'd like in early.

A recent redesign of our landing page quietly let a handful of labels and captions slip below our contrast standard. We caught it and put every piece of text on the page back to WCAG AAA (7:1) against its background - the same bar we hold the rest of ThreadScope to. Readable, accessible text isn't a nice-to-have here: if it's hard to read, we treat that as a bug and fix it.

The AI assistant can now reliably open and read the articles and links in a discussion - including JavaScript-heavy pages it used to choke on - so its answers and drafts are grounded in what the sources actually say. Suggested replies got sharper too: they credit the commenters who raised a point, build on the discussion instead of repeating it, stay gracious when disagreeing, and stick to tools and facts that are actually in the thread. We also fixed a bug where closing and reopening the chat could hide your most recent messages until you refreshed.

When a post just links to another Reddit post, we now pull that post in and show it inline - and the AI assistant can use it as context. The chat got a batch of upgrades too: suggested replies appear as clean, copy-ready blocks; the window grows as your conversation does; "Reset Chat" starts fresh context without losing your history; and links open in a new tab. On the dashboard, a new "All" tab keeps every post in view so nothing vanishes once you start working on it, and you can press j or k to jump to the next or previous post without going back to the list.

Alongside the free Lookout plan we're introducing Scout, Patrol and Command, priced by how many Reddit post analyses you get each month rather than by credits. You'll see the full lineup on the pricing page; some tiers are arriving shortly.

Switch plans, resume a cancelled subscription and see your live billing status without leaving ThreadScope. Upgrades take effect straight away; downgrades apply at your next renewal, so you keep the tier you've paid for until then.

On paid plans you no longer choose a model by hand. ThreadScope now selects the best model for the job automatically, so we can keep quality high for you.

Semantic tags now produce fewer false positives. Descriptions that rule a topic out ("...but not database chat") are respected, and borderline matches get a quick second-opinion check before a post is tagged.

Reddit posts and comments can contain text that tries to hijack an AI assistant. Chat now treats fetched content strictly as data and screens both that content and your own messages for injection attempts before answering.

If you added a subreddit that another ThreadScope user was already following, your existing posts were tagged by keyword only — semantic matches were skipped. ThreadScope now runs the same embedding match on those linked-in posts as on freshly fetched ones.

This panel. Open it from your account menu to see what's changed since the last time you looked. A dot on the menu means there's something new.

All text colours now meet WCAG AAA contrast (7:1) against both card and page backgrounds. Sharper, less squinting.

When you upgrade to a plan with semantic matching, your existing tags get LLM-generated descriptions automatically so semantic matching kicks in right away.

Polling intervals now adapt to each subreddit's activity. Busy communities get checked more often; quiet ones less.

Paid tier tags now match posts by meaning, not just keywords. Write a description of what you care about and the embedding model finds matches even when the words don't line up.

New admin tooling: a unified event log across LLM calls, tool calls, and system events, plus an AI assistant that can answer questions about it.

Bold, lists, code blocks, and tables in chat responses now render as you'd expect instead of showing raw markdown.

New accounts get a welcome email. Daily digest emails are opt-in from a banner on first login.

Tag and subreddit filters merged into a single control. Selections persist across page reloads. Sidebar is interactive again.

Pull post content, comments, and external links into AI chat as attachments. Started tab tracks posts you've drafted but not yet responded to.

Subreddit filter, default tag filter, and several layout glitches addressed on small screens.

Social login added. If you sign up with one method and later use another with the same email, accounts are merged automatically.

Connect OpenRouter via OAuth to get free-tier AI chat without managing API keys. Replaces the old Groq BYOK flow.

ThreadScope is live. Pick the subreddits that matter to you, group the terms you care about into tags, and triage new posts as they come in with keyboard-driven shortcuts. An AI assistant reads each post and its discussion so you can draft a thoughtful reply in seconds.