Developer builds deliberately broken MCP server to test contract verification tools

·2 views

A developer created 'mcp-worse', an intentionally deceptive Model Context Protocol server designed to violate documented API contracts by reversing tool order and omitting cache metadata. The project was built to demonstrate a core problem: MCP servers can claim statelessness, cacheable lists, and stable tool ordering without the protocol enforcing any of it. To validate a contract-checking tool called 'contrast-smoke', the developer needed a real binary that reliably fails in specific, predictable ways rather than a hypothetical scenario. The contrast-smoke test spawns both the honest server and the lying one as actual processes, inspects live wire traffic, and only exits successfully if the good server passes and the bad one fails. The project, published on GitHub, is labeled as a teaching tool and is not intended for production or registry deployment.

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)

Microsoft Copilot Tools Lose Momentum as Pricing and Pivots Deter Adoption

Microsoft launched three developer-focused AI tools — Code Apps, Cowork, and the new Copilot Studio — that initially generated strong enthusiasm for their productivity potential. However, adoption has since been hampered by incomplete rollouts, with Code Apps' citizen-developer interface still in preview nearly a year after launch, and Cowork shifting to a costly pay-as-you-go model. New Copilot Studio also began charging for development while removing free access for Microsoft 365 Copilot users, a move significant enough that Microsoft appears to be retaining the older version rather than replacing it. Critics argue these pricing shifts enforce a 'prove ROI before trying' mindset that stifles innovation and discourages citizen developers. The broader Copilot strategy has also faced criticism for fragmented rollouts and repeated architectural pivots, leaving the platform's early reputation weakened despite recent improvements.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Building a browser-only media converter with FFmpeg.wasm: deadlocks, limits and workarounds

A developer built BrowsersKit, a fully client-side media conversion tool using FFmpeg.wasm, to eliminate server uploads and protect user file privacy. The project encountered serious technical hurdles, including thread pool deadlocks in FFmpeg.wasm's multithreaded build that silently hung roughly one in ten heavy jobs with no error output. A watchdog mechanism was eventually shipped to detect and recover from these freezes. The 32-bit WebAssembly heap imposed a 2 GB memory ceiling, requiring a workaround using WORKERFS to handle larger files. Additionally, the developer chose to serve VP8 video instead of VP9 despite user requests, citing codec reliability issues within the browser environment.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Diagnose VPN DNS Issues by Identifying the Layer Causing the Problem

When DNS behaves unexpectedly, the VPN is often blamed, but multiple layers — including the OS, browser, router, app cache, and VPN profile — can each influence how a lookup is resolved. A structured diagnostic approach involves first identifying where the problem appears, such as whether it affects one browser, one device, or all devices on a network. Comparing behavior across browsers, devices, and VPN states — while changing only one variable at a time — helps isolate the responsible layer without disrupting a working setup. DNS caching means recent changes may not appear immediately, making a device restart or a timed retest a valid early step. Precisely describing the symptom, including when it started and under what conditions it occurs, is key to making the problem actionable.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

PostgreSQL XID Wraparound: How a Math Limit Can Halt Your Database

PostgreSQL uses a 32-bit Transaction ID (XID) system with roughly 4.2 billion available IDs, treating them as a circular ring to manage data visibility under its MVCC concurrency model. When old rows are not periodically 'frozen,' their transaction IDs can mathematically shift into the future, making valid data invisible to queries. To prevent data corruption, PostgreSQL forcefully blocks all new write operations once the counter comes within 11 million transactions of wraparound. Autovacuum normally handles row freezing in the background, but it can be silently stalled by abandoned replication slots, orphaned prepared transactions, or long-running uncommitted queries. Monitoring the datfrozenxid age and ensuring Autovacuum runs without blockers are the primary safeguards against this type of outage.

0 comments Read more at DEV Community