Radar, November 6, 2026: ten signals from X on Claude Finance, Cursor, Antigravity and MCP
In short: The standout signal of November 6: Claude gets Personal Finance in the US, a read-only connection to your own accounts. Alongside it, ten short signals on dev tools, MCP and ChatGPT security, each with its source on X. This edition was compiled afterwards, on November 8, 2026, from the X briefings of November 6.
Illustration by Claude (AI), drawn with code: a radar screen with ten signals.
1.Claude gets Personal Finance in beta (US)
Confidence: high (own announcement)Anthropic introduced Personal Finance in Claude, in beta and US-only. You connect accounts you choose and Claude answers questions about balances and transactions. The connection is read-only: Claude cannot move money. Decide beforehand which accounts you actually want to share.
Source on X: @claudeai
Context in the briefing: Personal Finance in the November 6 briefing
Read also: Claude Personal Finance explained
2.Cursor builds live apps you share with a link
Confidence: high (own announcement)In Cursor you describe a tool, then Cursor builds it, sets up a database, runs security checks and deploys; you share the app with a link. The one-click deploy runs on Vercel. For internal tools that saves many separate steps; do check yourself which data such an app can read.
Source on X: @cursor_ai
Context in the briefing: Cursor in the November 6 briefing
Read also: Skills and plugins
3.Antigravity gets remote agents in cloud VMs
Confidence: high (own announcement)Google Antigravity is rolling out remote agents this week: tasks run in separate cloud virtual machines and you approve permissions from your phone. A new interface and Jupyter notebook support arrived earlier. Long tasks no longer have to run on your own laptop.
Source on X: @antigravity
Context in the briefing: Antigravity in the November 6 briefing
4.GitHub Copilot Fusion picks the model itself
Confidence: high (own announcement)Copilot Fusion routes tasks across several models in three modes (Single, Cascade and Critique); you only choose the effort level. It works in the Copilot app, the CLI and VS Code, with a 30% discount until November 30. If you prefer to fix the model yourself, the regular mode keeps that choice.
Source on X: @github
Read also: Skills and plugins
5.Vercel Schedules: agents schedule their own tasks
Confidence: high (own announcement)Vercel Schedules is in public beta. Agents and apps can schedule their own tasks with it: fixed jobs are defined in code, dynamic jobs are created at runtime, previews included. For many small projects it replaces a separate cron server.
Source on X: @vercel
6.Devin Review 2.0 for pull requests: cheaper and faster
Confidence: high (own announcement; figures from the maker)Cognition released Devin Review 2.0 for pull request reviews. Cognition says it is up to five times cheaper and one and a half times faster, with new features and integrations. Those figures come from the maker; compare them with your own review time before switching.
Source on X: @cognition
7.Supabase: your own MCP server for your app, with OAuth 2.1
Confidence: high (own announcement)Supabase shows how to add an MCP server to your app from its own library, so customers can use your app through agents. Supabase Auth provides the OAuth 2.1 server that MCP expects for sign-in. For indie developers it is a short route to an app that agents can call.
Source on X: @supabase
Read also: MCP and agent tooling
8.Perplexity Teammates: proactive agents for Max subscribers
Confidence: medium (TestingCatalog report)According to TestingCatalog, Perplexity is rolling out Teammates, proactive agents for Max subscribers on the web. They run in an isolated environment with Computer tasks, skills, memory and tools; mobile follows later. Perplexity did not confirm this itself in this source.
Source on X: @testingcatalog
Read also: MCP and agent tooling
9.OpenAI Private Safety Processing: safety checks without storing prompts
Confidence: high (own announcement)With Private Safety Processing, OpenAI can check API requests for safety without storing prompts or responses. The records stay in your environment and an agent reviews them there. If you work with sensitive data, that may change the trade-off around data retention; read the terms yourself.
Source on X: @OpenAIDevs
Context in the briefing: Security in the November 6 briefing
10.ChatGPT Lockdown Mode could be bypassed through HTML previews
Confidence: medium (trade outlet, no OpenAI response in the source)According to The Hacker News, researchers found a flaw in HTML previews that allowed ChatGPT's Lockdown Mode to be bypassed, including zero-click attacks on the desktop app. Private chats could leak unnoticed. The post does not say whether the flaw has been fixed; keep the app updated.
Source on X: @TheHackersNews
