wayback-markdown MCP server
Commit Oct 25, 2026Lets AI agents browse Internet Archive snapshots in clean Markdown with metadata frontmatter and context-efficient pagination.
Also known as ertug/wayback-markdown
- Stars
- 1
- Tools
- 3
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 25, 2026
- License
- MIT
What is wayback-markdown MCP?
wayback-markdown lets AI agents browse Internet Archive snapshots in clean Markdown with metadata frontmatter and context-efficient pagination. It exposes 3 tools, such as list, get, links.
Summary generated from the project's README
About wayback-markdown MCP server
Browsing historical web snapshots through an AI agent is what wayback-markdown enables. This MCP server and CLI tool allows developers to access the Internet Archive by converting snapshots into clean Markdown. It is designed to be context-efficient for LLMs by providing a YAML frontmatter block containing metadata like capture time, redirects, and HTTP status, while truncating the body to a specific character budget to save tokens.
Users can connect this server to any MCP client, such as Claude Code, using the mcp subcommand. It handles various file types including HTML, text, PDF, DOCX, and PPTX via markitdown. The tool bypasses common restrictions that prevent AI agents from accessing web.archive.org by fetching raw endpoints directly. It also features a disk cache to serve identical requests quickly and rewrites links and images to ensure they point to archived versions rather than live sites.
What you can do with it
- 1Researching historical website changesAn agent can find a specific year's capture of a website and follow its archived links to explore old content.
- 2Comparing archived documentsYou can ask an AI to compare the homepages of two different eras or find specific shutdown notices from the past.
- 3Summarizing old file formatsAn agent can retrieve and summarize archived PDF reports or other documents from a specific point in time.
- 4Auditing historical site rulesA developer can use the tool to compare old community guidelines with modern versions to identify rule changes.
Written from the project's README. Every use case is checked against a quote from it.
wayback-markdown MCP tools 3
Detected from the READMElistgetlinks
Configuration
WAYBACK_MARKDOWN_CACHE
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachableertug/wayback-markdown is public and reachablegithub.com/ertug/wayback-markdown ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-10-25github.com/ertug/wayback-markdown ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model3 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubwayback-markdown
Let your AI agents browse the Wayback Machine in clean Markdown — an MCP server and CLI.
It fetches an Internet Archive snapshot, converts it to agent-friendly Markdown with markitdown (HTML/text plus PDF, DOCX, PPTX), and returns a metadata frontmatter + length-capped body — so an agent sees the signals that matter (capture time, redirects, status) without flooding its context.
[!NOTE] Please use it responsibly and avoid creating unnecessary load on the archive — the Internet Archive is a free, donation-funded nonprofit.
Why
- Agent-ready — connect it to any AI agent or MCP client with a single command.
- Context-efficient — spends few tokens per read: outputs Markdown (no HTML noise), leads with a YAML frontmatter block (capture time, redirects, status), truncates to a character budget, and pages the rest on demand, so the agent gets the signals up front.
- Toolbar-free — no injected Wayback UI leaks into the Markdown.
- Self-navigating — every link and image is rewritten to its archived form (honoring
<base href>). Links in the output are validgetinputs: paste one back in to keep browsing. - Cached — identical requests are served from a disk cache. A capture addressed by its full timestamp is cached forever (it's immutable); "latest"/nearest lookups refresh daily.
- Unblocked —
web.archive.orgis typically off-limits to AI agents (e.g. Claude Code's built-inWebFetchrefuses the host). This tool fetches the raw…id_/…endpoint directly overhttpx, so an agent can read captures anyway. - Redirect-aware — surfaces archive redirects (you got a different capture than asked),
crawl-time redirects (the page was a 3xx when captured), and client-side
<meta refresh>redirects. - Meta-aware — surfaces the
<title>anddescription/keywords/author<meta>tags markitdown drops. - Frame-aware — the frontmatter lists each
<frame>target as a readygetinput (including ones written at runtime bydocument.write). - Fallback-aware — re-parses
<noframes>/<noscript>bodies as markup, so a page whose real content hides in one converts properly instead of dumping unparsed tags. - Charset-aware — decodes each capture by its declared charset (HTTP
Content-Typeheader first, then<meta charset>/BOM, sniffing only as a fallback), so legacy Windows-1252/Latin-1 pages keep bytes like®instead of�.
Demo (CLI)
$ wayback-markdown get http://www.python.org --at 1997 --max-chars 500
---
requested-url: http://www.python.org
requested-timestamp: 1997 (1997)
served-timestamp: 19980119014227 (1998-01-19 01:42:27 UTC)
timestamp-note: no capture at the requested time; served the nearest one
final-url: https://web.archive.org/web/19980119014227id_/http://www.python.org/
http-status: 200
redirects:
- https://web.archive.org/web/1997id_/http://www.python.org
mimetype: text/html
title: "Python Language Home Page"
description: "Home page for Python, an interpreted, interactive, object-oriented, extensible programming language. It provides an extraordinary combination of clarity and versatility, and is free and comprehensively ported."
keywords: "Python programming language object oriented web free source"
markdown-chars: 11225
---
**[Home](https://web.archive.org/web/19980119014227/http://www.python.org/) |
[Software](https://web.archive.org/web/19980119014227/http://www.python.org/python/) |
[Documentation](https://web.archive.org/web/19980119014227/http://www.python.org/doc/) |
[PSA](https://web.archive.org/web/19980119014227/http://www.python.org/psa/) |
[Workshops](https://web.archive.org/web/19980119014227/http://www.python.org/w
[truncated: showing chars 0-500 of 11225 total. Continue from offset 500 for more.]
Install
Needs uv (curl -LsSf https://astral.sh/uv/install.sh | sh, or see the install docs):
uv tool install git+https://github.com/ertug/wayback-markdown.git
wayback-markdown --help
MCP server
Point any MCP client at the mcp subcommand, e.g. with Claude Code:
claude mcp add wayback-markdown -- wayback-markdown mcp
Or add it to an MCP client config directly:
{
"mcpServers": {
"wayback-markdown": {
"command": "wayback-markdown",
"args": ["mcp"]
}
}
}
Optionally, set the cache location by adding an "env" block — e.g. to keep it across
reboots: "env": { "WAYBACK_MARKDOWN_CACHE": "~/.cache/wayback-markdown" }.
Tools
list— find captures of a URL: date range, HTTP status, URL/MIME-type regex filters, andmatch=prefix|domainto sweep a whole path or site.get— fetch a snapshot as frontmatter + Markdown; picks the capture closest toat, truncates tomax_chars, and pages long documents viaoffset.links— list a snapshot's outbound links, each rewritten to its archived URL, ready toget.
The same tools are available as CLI subcommands (wayback-markdown list|get|links,
see --help) — as used in the Demo and Formats sections.
[!WARNING] Archived pages are untrusted third-party content. As with any web-fetching tool, the output may contain text that tries to manipulate the agent (prompt injection). The server's instructions do tell the agent to treat fetched content as page data, never as directions to follow, but don't rely on that alone — run the agent with least privilege and review what it does with fetched content.
Prompt an agent
With the server connected, hand the agent a prompt and let it list, fetch, and follow captures on its own — ideal for the obscure and the long-gone:
- "Google started at google.stanford.edu. Find its 1998 capture and walk its links to see what it offered."
- "Yahoo bought GeoCities in 1999. Compare geocities.com's homepage before and after the deal, then find the shutdown notice from 2009."
- "Enron collapsed in 2001. Find its last annual-report PDF archived on enron.com and summarize what it said about risk."
- "Find the archived Hacker News guidelines and FAQ from the early years, compare them with the latest capture, and say which rules were added."
Formats
Run it directly on one real capture per format — each doubles as a smoke test for its converter:
# HTML — the first website
wayback-markdown get 'http://info.cern.ch/hypertext/WWW/TheProject.html' --at 19990427150243
# text — robots.txt, passed through verbatim
wayback-markdown get 'http://www.google.com/robots.txt' --at 20010206202714
# PDF — the Bitcoin whitepaper
wayback-markdown get 'http://www.bitcoin.org:80/bitcoin.pdf' --at 20100704213649
# DOCX — IEEE's conference-paper template
wayback-markdown get 'https://www.ieee.org/content/dam/ieee-org/ieee/web/org/conferences/conference-template-a4.docx' --at 20200823052644
# PPTX — a NASA talk on the JWST mirror coatings
wayback-markdown get 'https://jwst.nasa.gov/resources/SPIE20128442-89RKeski-Kuha.pptx' --at 20130219022902
Configuration & development
$WAYBACK_MARKDOWN_CACHE(env var) — cache location (default/tmp/wayback-markdown-cacheon Linux).uv synconce, thenuv run pytest— offline unit tests, no network.
wayback-markdown MCP questions
What tools does wayback-markdown MCP provide?
It exposes 3 tools, including list, get, links.
How does it handle large web pages?
It uses a character budget to truncate the body and provides pagination via an offset so the agent can request more content on demand.
Is the data cached to save bandwidth?
Yes, identical requests are served from a disk cache, and captures addressed by a full timestamp are cached indefinitely.
Does it work with AI agents that usually block the Wayback Machine?
Yes, it bypasses typical blocks by fetching the raw endpoint directly over httpx.