time-machine-mcp
# time-machine-mcp
An MCP server that lets Claude show you how any website looked on any date, using the Internet Archive's Wayback Machine. No API keys required.
## Ask things like
- "What did google.com look like in 1999?"
- "How has bbc.co.uk changed between 2000 and 2010?"
- "When was apple.com first archived?"
- "Show me nytimes.com on the day after the 2008 election."
## Install
You need [uv](https://docs.astral.sh/uv/) and Python 3.11+.
```bash
git clone https://github.com/Prashan7h/time-machine-mcp.git
cd time-machine-mcp
uv sync
```
## Set up with Claude Desktop
Open your config file (Claude Desktop → Settings → Developer → Edit Config) and add:
```json
{
"mcpServers": {
"time-machine": {
"command": "uv",
"args": ["run", "--directory", "/path/to/time-machine-mcp", "server.py"]
}
}
}
```
Replace `/path/to/time-machine-mcp` with the full path to your clone, then restart Claude Desktop. If Claude Desktop can't find `uv`, use its full path (run `which uv` to find it).
## Set up with Claude Code
```bash
claude mcp add time-machine -- uv run --directory /path/to/time-machine-mcp server.py
```
## Tools
| Tool | What it does |
| --- | --- |
| `snapshot(url, date)` | Finds the archived snapshot closest to a date. Dates can be `YYYYMMDD`, `YYYYMM` or `YYYY`. |
| `history(url, from_year, to_year)` | Lists one snapshot per year, so you can watch a site change over time. |
| `first_seen(url)` | Finds the earliest archived snapshot of a site. |
URLs can be bare domains (`google.com`) or full URLs (`https://google.com`).
## Development
```bash
uv run pytest # unit tests (no network)
uv run pytest -m live # one live test against the real Wayback Machine
uv run mcp dev server.py # explore the tools in the MCP Inspector
```
## Credit
All the data comes from the [Internet Archive](https://archive.org)'s [Wayback Machine](https://web.archive.org), which has been preserving the web since 1996. This project isn't affiliated with them.
Please be kind to their free service: keep requests reasonable, and if you find this useful, consider [donating to the Internet Archive](https://archive.org/donate).
## License
MIT
TDQS
Scored across 3 tools
Each tool targets a distinct temporal query: nearest snapshot to a date, one snapshot per year for historical progression, and the earliest snapshot. There is no meaningful overlap in purpose.
Names are all lowercase and noun-oriented, but snapshot and history are single nouns while first_seen uses a snake_case descriptor. The set is readable and mostly consistent, though it lacks a uniform verb_noun pattern.
Three tools are well-scoped for a narrow web-archive lookup service. Each tool provides a unique query mode and none feels redundant.
The surface covers nearest-date lookup, yearly history, and earliest snapshot, which are the core retrieval needs. A way to list all snapshots within a range or explicitly fetch the latest snapshot is missing, though date-based workarounds exist.