Developer Builds MCP Server to Make Portfolio Queryable via Claude Desktop

·3 views

A developer named Ayush grew frustrated with static portfolio sites that display projects as mere lists of links and decided to build an interactive alternative. He created a Python-based MCP (Model Context Protocol) server using FastMCP that exposes nine of his projects as five queryable tools, allowing AI clients like Claude Desktop to fetch live, structured data on demand. Users can ask natural-language questions such as 'what has Ayush built with FastAPI?' and receive answers pulled directly from the server rather than from cached or guessed information. The project data is stored in plain Python dictionaries, keeping the setup lightweight with no database required. During development, Ayush encountered several setup hurdles, including path differences for the Windows Store version of Claude Desktop and the model sometimes answering from its own training data instead of calling the connected tool.

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)

Developer releases Gisia, a lightweight self-hosted Git and DevOps platform

A developer has publicly launched Gisia, an open-source, self-hosted Git and DevOps tool designed for personal and small-team use. The platform supports Git hosting over SSH and HTTPS, CI/CD pipelines defined in YAML, issue tracking, merge requests, and webhooks. Gisia is built to be lightweight, running entirely from a single Docker image that can be set up with a simple docker compose command. The tool also includes namespace runners for flexible CI runner management and group-based access control. The project is available on GitHub, and the developer is actively seeking community feedback to improve it further.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Rust Tutorial: How to Build Custom Iterators Using the Iterator Trait

A new entry in a Rust programming guide series explains how to create custom iterators by implementing the Iterator trait, which requires only a single method: next. The tutorial walks through building a Counter struct that iterates from 1 to 5 by incrementing an internal count field and returning None once the limit is reached. It then demonstrates a more advanced use case that chains multiple iterator methods — zip, map, filter, and sum — across two Counter instances to compute a final result of 18. Key iterator adaptor methods are explained individually, covering how zip pairs elements from two iterators, map transforms them, and filter selects only those divisible by 3. The guide is part of a broader chapter on functional programming features in Rust, including closures and iterator performance.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Five pipeline optimizations that slashed LLM token usage by 70% without quality loss

A development team found that rising API costs, inconsistent latency, and over-powered model usage were draining their LLM budget within weeks of deployment. Rather than simply downgrading to a cheaper model, they optimized the entire pipeline using techniques such as complexity-based routing, which directs simple tasks to lighter models and reserves expensive ones for advanced reasoning. They also trimmed bloated prompts, implemented semantic caching to serve repeated intents without re-calling the LLM, and filtered RAG context more aggressively to send only the most relevant document chunks. Adopting structured outputs further reduced unnecessary token generation and improved parsing reliability. Together, these changes cut token usage and API spend by roughly 70% while keeping response quality largely intact.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Lakers Roster Dashboard Using Claude to Surface Analyst Bias

A software developer used Anthropic's Claude Fable 5 model to build a Lakers roster-fit tool called 'Luka Fit Index' as a personal, no-stakes project. The dashboard evaluates new Lakers signings on four axes — spacing, play finishing, defensive cover, and ball-need — centered on Luka Dončić's league-leading 38.0% usage rate. Rather than producing a single composite score, the tool deliberately exposes its own assumptions and sample sizes, treating bias visibility as a core feature. A pre-build data check caught two key errors: Quentin Grimes shot a career-worst 33.4% from three last season, and the new starting center played only five games before shoulder surgery. These early corrections elevated confidence tags from minor footnotes to prominent UI elements that shifted assessments for nearly half the players evaluated.

0 comments Read more at DEV Community