Developer builds local-first MCP server to give AI agents accurate codebase architecture context

·2 views

A developer built CodeGraphy, a tool to help AI coding assistants understand the structure of large codebases. The problem it targets is that LLMs lose track of module connections as projects grow, wasting context on folder explanations or hallucinating imports. CodeGraphy is a local-first architecture visualizer that renders file dependencies as an interactive graph using React Flow, so no code is uploaded to the cloud. It also acts as a Model Context Protocol server, letting MCP-compatible clients such as Claude Desktop and Cursor query the real dependency graph before proposing architectural changes. The backend uses Python with FastAPI for AST parsing, and the tool is distributed as a desktop app and a lightweight Python package with a zero-config CLI.

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)

DocStar Lets AI Coding Agents Draft REST API Documentation for Review

Writing REST API docs by hand means copying routes, parameters and example bodies into a documentation tool, which takes days. DocStar shifts that work to AI coding agents such as Claude Code, Cursor or Codex, which read the codebase and generate draft entries covering method, URL, authentication, parameters and sample responses. Developers then review, test and publish an interactive reference, and the same documentation powers a hosted MCP server and a CLI. The guide outlines the components every endpoint needs, from purpose and headers to request bodies and error examples. It also explains how to connect the agent to DocStar's MCP server through a browser sign-in, with no API key required. Human reviewers should focus on naming, clear field descriptions, and keeping secrets out of examples.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Latchkey vs Blacksmith: GitHub Actions runner pricing and feature trade-offs compared

Latchkey and Blacksmith both swap ubuntu-latest for a managed runner with a one-line runs-on change and bill per minute, so the decision hinges on price, platform coverage, and how failed jobs are handled. Latchkey charges $0.0025 per minute for a 2 vCPU Linux x64 runner, compared with Blacksmith's $0.004, but Blacksmith offers arm64, Windows, and macOS runners plus 3,000 free minutes monthly, none of which Latchkey currently provides. Latchkey adds a self-healing agent that retries transient failures inside a run, while Blacksmith emphasizes cached dependencies and Docker layer reuse through its own infrastructure. The comparison uses each vendor's published rates and documentation as of 20 September 2026, and cells lacking published figures are marked as such.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

ChatGPT or Gemini in 2026? Each Wins at Different Tasks

A daily user of both ChatGPT and Gemini concludes that neither is better across the board, with the right pick depending on existing tools and typical work. Gemini suits people who already live in Google's apps like Gmail and Docs, and its paid tier includes generous storage. ChatGPT is generally stronger for natural-sounding writing, project-based work, and coding for learners, while Gemini's 1-million-token context window makes it better for analyzing very long documents. Both cost about $20 a month at the mainstream paid level, and pricing and model names change quickly, so readers should check official pricing pages before subscribing.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Apple's OLED iPad mini 8 brings a thinner body, brighter screen and A20 Pro chip

Apple has released the iPad mini 8, which it describes as the most significant refresh of the mini line in 14 years. The tablet is 1 mm thinner than its predecessor and weighs 255 g, making it the lightest iPad Apple has produced. Its display switches from LCD to OLED, offering deeper blacks and higher brightness, though it keeps a fixed 60 Hz refresh rate. The device runs the A20 Pro processor, the same chip used in the iPhone 18 Pro, and carries an IP67 water-resistance rating, a first for iPad.

0 comments Read more at DEV Community