AI Agents

MCP Station Report, November 3, 2026: Deprecated MCP Servers and a Day of Registry Changes

AI coding agents are installing MCP servers that no longer receive patches, according to a Help Net Security report published November 3, 2026. The report covers new research from Glow Labs, which reviewed real coding-session transcripts and calls the pattern MCPhantom.

Help Net Security reports that in 98% of the sessions Glow Labs examined, the agent appears to have filled in the package name from the model’s training data. In the other 2%, the user asked for the old package by name. The research looked only at sessions where a deprecated server was installed or suggested. According to the report, the pattern appeared across Claude Code, Cursor, Codex CLI, Claude Cowork and Claude chat, and across models from Anthropic, OpenAI and xAI.

The report says four high-severity vulnerabilities turned up in the deprecated reference servers. The maintainers closed Glow Labs’ report as out of scope because the repository was archived in May 2025. Help Net Security also says the deprecated packages Glow Labs counted have been downloaded from npm more than 17.3 million times since July 2025, and that no exploitation has been seen yet. In the report, Glow’s Nadav Claude Cohen said:

“A lookup before install is not enough on its own.”

What MCP Station measured

MCP Station tracks remote MCP servers listed in the Official MCP Registry. On November 3, 2026, the index held 37,055 servers. Of those, 32,529 were reachable when checked, or 88%.

Over the two-day period from November 2 to November 3, the registry showed a lot of movement:

  • 2,209 servers were newly listed.
  • 444 servers added tools and 105 removed tools. Across the registry, that came to 1,378 tools added and 539 removed.
  • 559 servers changed a tool schema, covering 1,864 tools.
  • 44 servers added new destructive tools, 89 tools in total.
  • 222 servers went down and 110 came back.

These measurements describe the current state of listed remote servers. They do not measure which packages coding agents install locally, and they do not show whether any listed server is deprecated. The Glow Labs findings concern packages run through npm, and the Station figures concern servers reachable over the network. We present them side by side as context. Neither is evidence about the other.

Why it matters for agents

The Glow Labs research concerns how agents choose servers. The Station figures show how quickly the registry changes: servers appear, tools are added or removed, schemas shift, and some servers go offline and later return. Help Net Security reports that Glow’s suggested first steps include approving one server per service and adding npm deprecation status to approval criteria. Both sets of facts point to the same practical question for agent operators. How current is the information an agent relies on when it selects a server?

Measured changes

32,529 of 37,055 remote MCP servers answered on the 2026-11-03 check.

Change (2026-11-02 to 2026-11-03) Servers
Newly listed servers 2209
Servers that added tools 444
Servers that removed tools 105
Servers with input schema changes 559
Servers with new destructive tools 44
Reachability: came back 110
Reachability: went down 222

Largest categories

Category Servers

Sources

About this data

Measurements by MCP Station, which checks every remote server in the Official MCP Registry daily. Measurements are not a security guarantee and are not an endorsement of any server. Methods: mcpstation.net/methods.html. Data: mcpstation.net/api.

Disclosure: the operator of Emerge Times also operates MCP Station.

This article was prepared with AI assistance and reviewed by an Emerge Times editor before publication. Licensed CC BY 4.0: credit “Emerge Times (emergetimes.com)”. Usage & License