Radware launched MCP Server Protection to give orgs visibility into MCP servers, tools, and agent activity, and to block protocol abuse, auth attacks, tool/resource enumeration, unauthorized interactions, AI-driven tool flooding, and DoS. It runs as part of Radware’s cloud app protection or as a cloud-augmented add-on for on-prem Alteon and DefensePro.
In an Oct 14 OpenAI case study, Amp CEO Quinn Slack says OpenAI models make up the majority of tokens used in Amp. He adds that some users burn 10B tokens a week, and that Calling Puck, built on GPT Live 1, lets you talk in real time, even from a watch, to the Amp assistant that launches and coordinates your other agents…
Origin, Cursor's code hosting platform, now plugs into 20+ apps including Vercel, Cloudflare, CircleCI, CodeRabbit, Greptile, Semgrep, Sentry, and Linear, and any dev tools team can now submit its own. Apps are built against the Origin API (app JWT, webhooks, check runs, PR comments), made public, then submitted with a demo video for Cursor's review…
HydraFusion’s research preview now draws from Claude Opus 5.5, GPT 6 Astra, GPT 6.1 Sol, and GPT 6 Luna, tailors reasoning effort per stage, and upgrades Cascade with a judge/repair step and Critique with parallel tool using critics. New HydraFusion Max targets harder tasks; both options stay in preview on Copilot Pro through Enterprise in VS Code, the Copilot app, and CLI…
Pantheon opened a public beta for its MCP server so agents in Claude Code, Cursor, and Copilot can list sites and environments, pull build and runtime logs, spin up Multidevs, and trigger rebuilds or rollbacks under your existing Pantheon account permissions. Read access is on by default; any state changing action pauses until you approve it in the client…
Claude Code 2.1.300 adds a prompt.change mod hook for non typed prompt edits (history recall, clear, undo, plugin fill), $.ui.view({ agentId }) to show an agent transcript in the main view, and OpenTelemetry GenAI semantic attributes on trace spans. It also splits cache write metrics into 5 minute and 1 hour buckets and lets agent.spawn rewrite workflow agent prompt and model fields…
Spotted something we missed? Start a thread.