Query Wayback Machine archives — search captures, retrieve snapshots, compare page versions, check archival status via mcp-wayback-machine (npx)

mcp-wayback-machine v3.7.1 — 8-tool Wayback Machine MCP server

Install: npm install mcp-wayback-machine → entry point dist/bin.js (stdio transport, no args needed)

Tools (8)

Verified execution — 15 calls, 15/15 success

Test matrix:

1. health → {"status":"ok","version":"3.7.1"} (4ms)
2. check_archive_status({url:"https://example.com"}) → HTTP 498 from archive.org CDX API (12343ms)
3. check_archive_status({url:"https://google.com"}) → HTTP 498 (11484ms)
4. search_archives({url:"https://example.com", limit:5}) → 5 results, earliest 2002-01-20, web.archive.org URLs (19948ms)
5. search_archives({url:"https://nytimes.com", from:"20200101", to:"20200131", limit:3, filter:["statuscode:200"]}) → 3 results, Jan 2020 only (6477ms)
6. search_archives({url:"https://example.com", collapse:"timestamp:8", limit:5}) → 5 unique-per-hour captures (22480ms)
7. search_archives({url:"https://example.com", matchType:"domain", limit:3}) → domain-wide results (449ms)
8. get_archived_url({url:"https://example.com", timestamp:"latest"}) → full HTML of example.com, timestamp 20260630040541 (3853ms)
9. get_archived_url({url:"https://example.com", timestamp:"20150101120000"}) → 2015-era HTML, resolved to 20150101120147 (5141ms)
10. get_archived_url({url:"https://example.com", timestamp:"latest", modifier:"id_"}) → raw content, cached from call 8 (3ms)
11. list_screenshots({url:"https://google.com", limit:3}) → 30s timeout (30017ms)
12. compare_snapshots({url:"https://example.com"}) → compared 2026-06-30 03:05:12 vs 04:05:41, includes visual diff URL (21184ms)
13. check_archive_status({url:"https://thisdomaindoesnotexist12345678.com"}) → HTTP 498 (11493ms)
14. search_archives({url:"https://example.com", filter:["!mimetype:image.*"], limit:3}) → negative filter works (16055ms)
15. clear_cache({}) → "Cache cleared successfully" (5ms)

Key gotchas

• ⚠️ `check_archive_status` returns HTTP 498 — archive.org's Spark API endpoint appears unstable/rate-limited; search_archives via CDX API is the reliable alternative
• ⚠️ `list_screenshots` TIMES OUT (30s default) for popular sites — archive.org screenshot index is slow
• `search_archives` is the workhorse — CDX API is reliable, supports matchType (exact/prefix/host/domain), date ranges, collapse (dedup by timestamp granularity or digest), filter (regex on fields, ! to negate), pagination
• Timestamp format is YYYYMMDDhhmmss (14 digits) — "latest" string also accepted by get_archived_url
• `modifier` enum: `id_` (raw), `im_` (screenshot), `js_` (JavaScript), `cs_` (CSS) — default id_ strips Wayback toolbar
• Content wrapped in security boundary — responses include --- BEGIN UNTRUSTED ARCHIVED CONTENT --- markers
• Built-in caching — repeated calls return instantly (3ms vs 5141ms); use clear_cache after save_url
• `compare_snapshots` auto-selects timestamps if omitted — defaults to oldest vs newest available
• Network-bound latency — p50=11484ms (archive.org CDX API), local operations (health, clear_cache) are instant
• **No API key needed for read