Because the corpus is your own docsets rather than a web index, the answer matches the version in front of you instead of whatever ranks well today, and nothing about the query leaves the machine. Two limits are worth knowing up front: full-text search is per-docset and off until you enable it, and both the docset list and the search results are truncated when they would exceed 25,000 tokens.
A server over the Dash docsets installed on your Mac, searching them and returning pages through Dash's own local API.
- The installed docsets listed with the identifiers the other tools address them by
- Search across a chosen set of docsets, with or without code snippets, capped by a result limit you set
- Full-text search switched on for a docset when matching on titles is not finding what you need
- A result opened as plain text with markdown-style links, loaded from the URL the search returned
macOS with Dash 8 installed and the docsets you intend to search already downloaded — the server reads what is there, and the listing comes back empty when nothing is. Python 3.12 or newer and uv to run it. Page loads only accept a URL pointing at the local Dash API on 127.0.0.1. MIT.
One command — uvx --from git+https://github.com/Kapeli/dash-mcp-server.git dash-mcp-server