Starting today, the AuthPlane x Skybridge Speedrun Challenge asks developers to build a real MCP App with Skybridge (the open source React framework for MCP Apps) and secure it with AuthPlane. The ask: a five-minute video showing it working. The prize: $500. The window: two weeks, closing Aug 31. Full rules: champ.ly/tUZ0lKNQ

Building an MCP App with Skybridge is the easy part. Adding authentication on top used to be the annoying part. We're teaming up with AuthPlane to see how fast that can be too. $500 prize. More soon.

A malicious MCP server was asked where to send the user to log in. It answered with something the operating system ran instead. JFrog disclosed this in July 2025. It scored 9.6 out of 10 on the standard severity scale, which is about as bad as these get. mcp-remote is the proxy that connects clients like Claude Desktop and Cursor to remote servers. On connect, it asks the server for its OAuth metadata, including the address to open in a browser for login. A malicious server returns a value that is not an address. The proxy hands it to the operating system, and on Windows it executes. Command execution on the developer's machine, before a single tool ran. Almost all MCP security work right now points at the tool layer. Meanwhile every client still takes an unknown server's word for how authentication is supposed to work. Metadata from a server you do not trust is untrusted input. JFrog's full write-up, including the technical chain: champ.ly/yLoqeIf- CVE-2025-6514, affecting mcp-remote 0.0.5 through 0.1.15. If you are on an affected version, update to 0.1.16 and connect only over HTTPS. #MCP #OAuth #AppSec #AIAgents

Nobody’s MCP compliance badge went stale last week. Including ours. The 2026-07-28 release added a formal deprecation policy for the first time: three lifecycle states, a public registry of deprecated features, and at least twelve months before anything deprecated can be removed. Ninety days is the floor, and only for an active security advisory. That matters because a version number was never the useful signal. What is compliant, what is deprecated, and what is not implemented yet are three different answers. Most badges collapse them into one. Ours is a public conformance catalogue, not just a badge. Every test case is readable, so you can check us instead of taking our word for it. MCP deprecation policy: champ.ly/WcLg_NA8 AuthPlane conformance catalogue: github.com/AuthPlane

The MCP 2026-07-28 spec finalizes today. Stateless core gets the headline, but 6 SEPs quietly rewire authorization too, and only one of them lands on the authorization server itself. We broke down what's changing and where AuthPlane stands against each one: champ.ly/capOAYEe #MCP #OAuth #AuthorizationServer #AIInfrastructure

Our new website is live: authplane.ai. Everything you need to understand and deploy open-source OAuth 2.1 for MCP, from core concepts to quickstart, all in one place. #AuthPlane #MCP #OAuth #AIAgents #OpenSource

Great event by @GitHub last night on the future of developer tools: luma.com/FutureCodeFrontier One point from @hahnbeelee stuck with us: as more code gets written by AI, more security incidents will follow unless builders get much better at understanding risk and mitigation. AI is changing how we ship. Security has to change just as fast. Panel: @idangazit @digitarald @dakshgup @hahnbeelee @gcgarriga

The MCP authentication problem has two layers. 🧵

The agent ecosystem is repeating mistakes the web already paid to fix. That's why we built AuthPlane: open-source OAuth 2.1 for MCP. github.com/authplane/authser… #MCP #AIAgents #OAuth #Security

This is a useful way to frame it. Once an MCP server can act on files, services, or credentials, public exposure becomes an execution/governance problem, not just a data leakage problem. Inventory + least privilege + revocation are going to matter a lot here.

Today we're launching AuthPlane. An open-source OAuth 2.1 authorization server purpose-built for MCP. Up and running in less than 10 minutes with a single Docker run command. Repo 👇 bit.ly/AuthPlaneX 🧵

AuthPlane sits between MCP clients (Claude Desktop, Claude Code, custom agents) and your servers. It brokers identity, consent, and token issuance, turning a security gap into production-grade auth in <10 min.

And when something breaks? No record of which agent did what, on whose behalf. IT can't see these connections, let alone govern them. The agent ecosystem is repeating every auth mistake the web spent 15 years fixing.

Most agents in prod run on a static API key copied into a .env file. One shared key. Never rotated. Never scoped. Never audited. One leak exposes everything. A compromised agent has god-mode.

But agents aren't toys anymore. They're calling real databases, email, file systems, and internal APIs. No auth = unauthorized data access, privilege escalation, data exfiltration, and unintended actions firing in prod.

The state of MCP auth right now: → 50% of MCP servers have no auth at all → 100% of the rest have at least one auth flaw Adding auth is error-prone and tedious, so it gets shipped as an afterthought.