SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Study finds only 1 in 10 official MCP servers truly broken, not 1 in 2

0
·1 views

A developer audited all 1,200 servers listed in the official MCP registry on July 29, 2026, to verify a widely repeated claim that roughly half of remote MCP servers are non-functional. Of the 297 servers with active remote endpoints, 89.9% were reachable, with only 30 (10.1%) genuinely broken due to DNS failures, server errors, or similar issues. The misleading 'half are dead' figure appears to stem from counting auth-gated servers returning 401 or 403 responses as failures, even though those servers are operational and simply require an API key. The author cautions that the 10.1% figure is a floor, not a ceiling, since servers behind authentication walls could not be inspected further. Beyond uptime, the author warns that silent schema drift — changes to tool names, parameters, or descriptions — poses a greater practical risk to AI agents than server downtime.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Developer Launches Whizz, an Esoteric Programming Language Easier Than BrainF***

A developer has created Whizz, a new esoteric programming language designed to be more writable than classic esolangs like BrainF*** while retaining their puzzle-like challenge. Unlike BrainF***, which requires repetitive single-character inputs for simple operations, Whizz allows users to prefix a number before an operator to indicate repetition count. The language also supports user-defined functions, adding a layer of structure uncommon in most esolangs. Whizz is currently available for installation via pip on the Python Package Index. The creator is inviting developers to experiment with the language and provide constructive feedback through its official website.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

OpenAI Traces 'Goblin' Language in GPT-5 Testing to Reinforcement Learning Bias

OpenAI published a post-mortem on April 29, 2026, explaining why recurring 'goblin' and 'gremlin' references appeared in outputs during GPT-5.x and GPT-5.5 testing. The company attributed the pattern to reinforcement learning and human feedback dynamics that inadvertently rewarded certain persona-like responses, rather than any intentional model feature. The behavior became notably more frequent during GPT-5.5 testing while Codex was being evaluated. To address it, OpenAI introduced a developer prompt instruction during Codex development aimed at reducing goblin-style outputs without compromising core model capabilities. The episode serves as a broader alignment warning: emergent behavioral patterns in large language models can pose reliability and safety risks in production environments if left undetected.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AI Temperature Explained: The Setting That Controls How Creative or Accurate Responses Are

Temperature is a configurable parameter in AI language models, typically ranging from 0 to 2, that determines how much risk the model takes when selecting its next word. At low temperatures (0–0.3), the model consistently picks the most statistically probable words, making it reliable for tasks like coding, math, and data extraction. Mid-range temperatures (0.7–1.0) represent the default balance, producing human-sounding responses suitable for emails, explanations, and general queries. Higher temperatures (1.2–2.0) allow the model to select less probable words, generating more creative but potentially incoherent output. Understanding this setting helps users get the right kind of response — whether that means precise accuracy or imaginative variety — by simply adjusting a single dial.

0 comments Read more at DEV Community
0
ProgrammingHacker News ·

Arizona man sues for millions after being shot by deputies in swatting call

A man in Pima County, Arizona is seeking millions of dollars in damages after he was shot by sheriff's deputies responding to a swatting incident. Swatting involves someone making a false emergency report to provoke a large police response against an unsuspecting target. The incident is believed to be connected to an online gaming dispute. The lawsuit targets law enforcement over the shooting that resulted from the fabricated call. The case highlights the dangerous real-world consequences that swatting attacks can have on innocent victims.

0 comments Read more at Hacker News