What it does
The server connects MCP clients such as Claude, Cursor or other agents to the Archive of Our Own (AO3). Its search_works tool supports filters and AO3's search-operator syntax, find_tags maps fuzzy wording to canonical tags, and get_work returns a metadata card for one work. read_works has a secondary Gemini model read 1 to 20 full fics and return a structured report for each, covering plot, characters, style, verbatim prose samples and content notes, plus a ranking against the user's question. The agent itself never reads the fic text, and fic downloads are cached locally for 24 hours.
It suits anyone who wants recommendations based on the actual text rather than the author's blurb. Writers can also use it for inspiration mid-scene, pulling reports on how other fics handle a similar beat, or for roleplay, using reports on fics with a strong character voice as style reference.
Running it requires Python 3.10 or later and a free Gemini API key, supplied as a launch argument or environment variable. Launch parameters set the reader model, a backup model and the minimum interval between AO3 requests, which defaults to 0.6 seconds. AO3 has no API, so the server scrapes its HTML one request at a time and honours Retry-After. The code is small and MIT-licensed, and the publisher intends it to be forked and modified.
Limitations and requirements
Requires Python 3.10 or later and a free Gemini API key. AO3 blocks plain HTTP clients; the server uses a browser TLS fingerprint, and if requests fail with 403 and a Cloudflare challenge, the IMPERSONATE setting in ao3.py must be changed. Fic text is sent to Google's Gemini API for reading. AO3 hosts works of all ratings, and the server passes through whatever the search returns, so the rating filter and AO3 warning tags should be used to control what is fetched.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
io.github.ArturLys/ao3-mcp
Search AO3 and have fics read by a secondary model before they're recommended
v0.1.2As shown on registry.modelcontextprotocol.io when we last read it