Model Context Protocol lets AI apps share and reuse tools across projects

·1 views

Model Context Protocol (MCP) enables AI applications to access shared catalogs of tools hosted on MCP servers, rather than each app defining its own tools independently. MCP servers, maintained by SaaS providers or internal AI platform teams, publish named, typed, and described tools that client applications retrieve and pass to AI models. Communication between clients and servers follows the JSON-RPC 2.0 protocol, transmitted over either streamable HTTP or stdio. Notably, the AI model itself never communicates directly via MCP — that role belongs to the surrounding tooling layer. Building an MCP server around existing tools is straightforward, as demonstrated using Python's FastMCP library to expose payment-related functions over a local HTTP endpoint.

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)

How a Content Pipeline's Batch Runner Silently Misreported Completion for Months

A batch runner built for a content pipeline on 16 March 2026 included resumability features from day one, yet the system remained functionally broken for three more months. The core problem involved four distinct bugs that each caused the runner to falsely believe work was already finished, including a quota error that returned an empty result instead of raising an exception, leading the pipeline to record missing content as successfully produced. Three separate parts of the stack independently handled rate-limit errors using different string-matching logic, meaning the same error could be caught or missed depending on where it surfaced. The team also had to manage stale 'running' states left behind by killed runs, as well as separate ownership of entity state and lock files — a distinction discovered only after a failure in production. The fixes for all four bugs landed in a single day, underscoring how resumability in batch pipelines differs fundamentally from retry logic and requires explicit state design.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Weedout Safari Extension Filters AI-Labeled YouTube Videos for $1.99

A developer has released Weedout, a $1.99 Safari extension for macOS designed to hide YouTube videos tagged with the platform's own 'Made with AI' label. The tool filters out such content from feeds, search results, related videos, playlists, and Shorts. The developer built it in response to a personal frustration with AI-generated conspiracy videos flooding their YouTube feed. Weedout processes everything locally and relies solely on YouTube's existing labels, meaning unlabeled AI content is not affected. The source code has been published on GitHub for developers who wish to create independent forks, though pull requests are not accepted.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

How Fancy Fox Built a Weather-Aware AI Recipe Pipeline Using FastAPI

The team behind Fancy Fox developed an automated daily recipe pipeline that factors in weather and seasonal context to generate AI-assisted recipes using FastAPI. Each pipeline run follows discrete steps — from fetching date and weather data to generating, validating, and publishing a structured recipe with an accompanying image. Model outputs are treated as untrusted input and parsed into strict Pydantic models, with custom validators normalizing ambiguous placeholder values like 'none' or 'n/a' to Python's None at the data boundary. The pipeline also manages browser sessions for social media promotion and uses rotating log files to retain enough history for debugging without consuming excessive storage. The public archive has grown to over 1,500 recipe pages, and the team has released a CC0 sample of 30 recipes and images for open use.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Open-Source AI Agent Tests Self-Rewriting Prompts, Rejects 'Mostly Right' Improvements

AgentSelfEdit is an open-source tool that allows an AI agent to rewrite its own system prompt based on execution feedback, then A/B tests edits to promote only statistically proven improvements. The project was tested on a 26-task classification benchmark where the baseline prompt scored just 46%, with the model making systematic errors around urgency detection, keyword over-indexing, and multi-label classification. An LLM-proposed edit added four priority rules, expanding the prompt from 212 to 939 characters, and correctly fixed four of the failing tasks. However, the edit broke one previously correct task, causing the system's quality gate to reject the change despite the net improvement. The project highlights a core challenge in self-improving AI systems: partial correctness is treated as insufficient, since any regression — however small — can disqualify an otherwise beneficial update.

0 comments Read more at DEV Community