v0.1.29

• Renamed suggest_intervention → suggest_review (breaking). The tool proposes a review for evipedia to write, so the old name described its argument rather than its effect. Arguments and behaviour are unchanged — suggest_review(intervention, goal?, references?, email?) still POSTs to the same public suggestion form. No alias is kept: MCP clients read the tool list on connect, so a reconnect picks up the new name

v0.1.28

• Added list_updates(days?) — the catalogue's change feed, newest first, backed by evipedia.ai's new /updates.json. Returns {title, slug, status, date} per entry, where status is new (first publication) or updated (an existing review revised). Called with no arguments it returns the last 7 days, capped at 100 entries (the raw feed spans the whole catalogue — 600+ rows, ~15k tokens, nearly all of it months old); pass days for a wider window. Answers recency questions — "what's new on evipedia?", "anything updated this week?" — that previously had no tool

v0.1.26

• search_reviews / GET /search results now report the full canonical topic verbatim (e.g. "Vitamin D for Health & Longevity", "Low-Level Light Therapy for Skin Rejuvenation") instead of the bare intervention name, so multiple reviews of the same intervention are distinguishable and the shape matches the evipedia site's search API. Dropped the internal category field. Each hit is now simply {topic, url}

v0.1.23

• Made search_reviews (and the REST GET /search) discovery-only: they now return the matching reviews (name + URL, ranked by relevance) so you can tell whether a review exists, instead of embedding each match's full conclusion. Results are ~15× smaller for broad queries; read a specific review with get_conclusion/get_review (MCP) or https://evipedia.ai/{slug}.md

v0.1.22

• Hosted server (mcp.evipedia.ai): added permissive CORS headers (with OPTIONS preflight) so browser-based clients can call the REST GET /search and the /mcp endpoint cross-origin, and added lightweight per-IP rate limiting (default 60 requests/minute) to /search and /mcp. Health checks and the stdio package are unaffected

v0.1.21

• Rebuilt search_reviews (and the hosted GET /search) on Lunr, configured identically to the evipedia.ai homepage search — same fields, boosts, and ranking, so the MCP tool, the REST endpoint, and the website return consistent results. Fixes natural-language queries that previously returned nothing: e.g. "low-level light therapy for skin rejuvenation" now finds the Skin review, and "vitamin d" ranks Vitamin D first

v0.1.20

Release 0.1.20

v0.1.18

• Added a hosted remote MCP server over Streamable HTTP at https://mcp.evipedia.ai/mcp — web-based and non-terminal MCP clients can connect with just a URL, no local Node/npx install needed. It serves the same read + suggest tools as the stdio package

v0.1.15

• Removed the internal env-var overrides (EVIPEDIA_BASE_URL, EVIPEDIA_SUGGEST_ENDPOINT) from the MCP Registry listing — they are optional dev-only knobs users never need to set
• Restructured the README headings (Tools, Install, Try it, Architecture › Public API Surface) and simplified the example config to bare evipedia-mcp (no @latest — npx already resolves the latest version)

v0.1.14

• Enriched the MCP Registry listing with a display title (“Evipedia”), websiteUrl, and icons (SVG + 192/512 px PNG served from evipedia.ai)