mcp-git

--- description: 'APPLY when working with mcp-http-fetch package TO understand package structure, architecture, available tools, and development guidelines' globs: alwaysApply: true --- # @ver0/mcp-http-fetch – Overview Rule ## Context - This rule is injected whenever files under `packages/mcp-http-fetch` are referenced. - It gives the AI a quick mental model of the package so it can navigate and edit confidently. - Use it for orientation; it is not exhaustive documentation. ## Core Idea `@ver0/mcp-http-fetch` is an **MCP (Model Context Protocol)** server exposing HTTP fetch operations as tools over **stdio**. It provides content-optimized fetching with **HTML minification**, **markdown conversion**, and **content limiting** for efficient AI processing. ## Key Directories ```text src/ index.ts // Server bootstrap + tool registration types.ts // Shared type definitions tools/ // One file per HTTP tool (fetch-html-tool.ts, fetch-markdown-tool.ts, ...) ``` ## Implemented Tools fetch-html · fetch-markdown · fetch-json Each tool follows the same pattern: 1. Declare an `inputSchema` with Zod. 2. Execute HTTP fetch with native `fetch` API. 3. Process content (minify HTML, convert to markdown, etc.). 4. Return structured response with metadata + content. 5. Include comprehensive error handling. ## Development Guidelines - Keep changes minimal & *pure-function* oriented. - Validate **all** inputs with Zod schemas. - Throw clear, actionable errors. - Process content efficiently for AI consumption. - Respect project-wide code quality & comment standards. ## Gotchas - All tools return metadata as first content item (JSON string). - Content processing (minification/conversion) may fail gracefully. - `start`/`end` parameters work on final processed content.