A small docs MCP server: search, retrieve, and keep track of the source
An agent can remember an API that no longer matches the project in front of it. Giving it documentation helps, but the source matters: the newest documentation and the documentation for an installed version are not necessarily the same. I use two small TypeScript servers to make that distinction explicit: tailwind-docs-mcp reads Tailwind's documentation source, while hermes-docs-mcp prefers a…
Two small TypeScript servers, named tailwind-docs-mcp and hermes-docs-mcp, have been developed to distinguish between the newest documentation and the installed version. These servers provide search, retrieval, and listing functionalities over MCP stdio. Neither server requires an embedding service or vector database. The servers parse the documentation, rank useful pages higher than incidental mentions, and maintain an up-to-date index.
Each server has three tools: Tailwind, Hermes, and Find candidate pages. The search returns titles, slugs, source URLs, and snippets. The search scorer prioritizes title and slug matches over body mentions and recognizes utility families. The retrieval process involves getting a specific page using the selected slug. Listing helps when the agent does not know the right terminology, and Hermes supports filtering by section.
The servers are built from pinned source using Git, npm, Node.js 20 or newer, and an MCP client that can launch stdio servers. The servers have been tested and passed, ensuring their functionality.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
