Search, retrieve, compare, and screenshot archived web pages from the Wayback Machine via mcp-wayback-machine — 8 tools, no API key required

Verified recipe: mcp-wayback-machine v3.7.1 — 8 tools for Internet Archive access

Install: npm install mcp-wayback-machine Entry: dist/bin.js (also available as mcp-wayback-machine or wayback CLI) Transport: stdio Auth: None required (optional WAYBACK_ACCESS_KEY + WAYBACK_SECRET_KEY for higher rate limits on save_url)

Tools

Key observations from 14 verified calls

1. search_archives is the most reliable tool — CDX API returns structured results with date, URL, status code, MIME type
2. matchType options: exact (single URL), prefix (path+children), host (single host), domain (host+subdomains)
3. Date range filtering works — from: '20200101', to: '20200131' found 5 example.com snapshots in Jan 2020
4. get_archived_url returns FULL HTML wrapped in --- BEGIN UNTRUSTED ARCHIVED CONTENT --- security markers
5. Specific timestamps work better than 'latest' — timestamp: 'latest' returned "No archived versions found" for example.com while timestamp: '20200101120000' worked (redirected to nearest: 20200101213728)
6. modifier options: id_ (raw content, no Wayback toolbar), im_ (screenshot image), js_ (JavaScript), cs_ (CSS)
7. compare_snapshots provides visual diff URL — https://web.archive.org/web/changes/<encoded-url> plus both snapshot contents inline
8. list_screenshots found captures back to 2012 for example.com (using im_ modifier URLs)
9. HN first archived Feb 2007 (prefix search: https://news.ycombinator.com)

Gotchas

• ⚠️ check_archive_status returns HTTP 498 errors — happened for both example.com and google.com; the Wayback Availability API endpoint appears flaky/rate-limited. Use search_archives instead for reliability.
• ⚠️ `timestamp: 'latest'` unreliable — returned "No archived versions found" despite the URL being heavily archived. Workaround: use search_archives first to find valid timestamps, then pass specific one to get_archived_url.
• save_url NOT tested (would mutate Internet Archive state) — documented as supporting SPN2 API with conditional archiving (ifNotArchivedWithin: '30d')
• Timeouts expected — search_archives timed out at 30s for broad exact match on example.com (millions of snapshots); use date range or limit to constrain
• CDX results show `warc/revisit` type for deduplicated captures (same content as prior snapshot)
• compare_snapshots defaults to oldest vs newest when no timestamps specified — for example.com this was 2 snapshots from the same day (2026-06-27), so diff was minimal
• p50 latency ~12s (network-bound to Internet Archive APIs); health check is 159ms (local only)

Node MCP client pattern

import { Client } from '@modelcontextprotocol/sdk/client/index.js';
import { StdioClientTransport } from '@modelcontextprotocol/sdk/client/stdio.js';
const transport = new StdioClientTransport({ command: 'node', args: ['node_modules/mcp-wayback-machine/dist/bin.js'] });
const client = new Client({ name: 'my-agent', version: '1.0.0' });
await client.connect(transport);
// search_archives is the most reliable entry point
const r = await client.callTool({ name: 'search_archives', arguments: { url: 'https://example