GitHub MCP is going viral, but most public MCP servers still lack strong auth: why it is not safe to use

What happened

GitHub MCP is currently one of the most talked-about MCP skills for people using Claude Code and Codex. In a high-engagement Reddit thread on August 6, 2026, users listed GitHub MCP among the servers they actually find useful in daily work.

At the same time, an August 11, 2026 security tracker summarized an Exposed by Design study that found 91.8% of audited MCP servers were running without OAuth. OAuth is a standard sign-in system that lets an app ask for limited access without handing over your main password. The tracker also said more than 21,000 internet-facing MCP instances were detected.

Anthropic's own guidance points users toward a curated marketplace of pre-approved MCP servers instead of random public integrations.

What it means for you

If you use an AI agent with a GitHub-related MCP skill, popularity does not guarantee safety. A third-party server can still be weakly protected, especially if you install the wrong copy or approve broad access. That could expose repository, issue, or release information to a server you did not mean to trust.

What to do instead

Use MCP skills from reviewed marketplaces or trusted vendors, not links passed around in forums. Check who runs the server before connecting it. Prefer limited permissions rather than full account access. If an integration asks for more access than you expect, stop and verify.

AgentPod lists only reviewed, tested skills, which can help reduce risk, but it is still worth checking what data a tool can reach before you turn it on.

Sources:

• https://www.reddit.com/r/ClaudeAI/comments/1vh0yd3/top_15_mcp_servers_that_are_actually_useful_in/
• https://www.deepinspect.ai/blog/mcp-security-news-tracker
• https://resources.anthropic.com/hubfs/Scaling%20agentic%20coding%20across%20your%20organization.pdf?hsLang=en