Free tool
The Xither record, inside your AI assistant
Xither runs a Model Context Protocol (MCP) server. Add it to Claude, ChatGPT, Cursor or VS Code and ask where a vendor publishes its sub-processor list, what notice window its data processing agreement states, or which models it retires next. Each answer comes from a stored reading and carries the page’s URL and the date Xither read it. No account, no key.
Endpoint
https://www.xither.com/mcp
Streamable HTTP. Read-only. 60 tool calls per 10 minutes from one network address. What you ask is not written to our database; the rate limit counts calls per address in memory.
Four tools
| Tool | What it returns | Limit |
|---|---|---|
| lookup_vendor | Look up one vendor's record | One vendor per call. Changes appear 30 days after Xither reads them. |
| search_vendors | Find vendors by name | Up to 10 names, slugs and record URLs. |
| notice_windows | Sub-processor notice windows | Only windows Xither can quote from the vendor's own terms. |
| upcoming_model_retirements | Upcoming model retirements | 90 days ahead by default, 365 at most; up to 50 rows. |
lookup_vendorreturns the same record as the vendor’s page on /vendors: the watched pages with their read dates, the pages Xither cannot read and why, the quoted notice window, public changes with permalinks, and dated model retirements. notice_windows is the notice windows table. Retirements come from the retirement tables vendors publish, as listed on model and API retirements.
Changes reach the server 30 days after Xither records them; subscribers get them the same morning, with their deadlines and a drafted notice (pricing).
Add it to your assistant
Claude (claude.ai and the desktop app)
Settings, then Connectors, then Add custom connector. Name it Xither and paste the endpoint URL. Leave the authentication fields empty. On Team and Enterprise plans an owner may need to add the connector for the organization first.
Claude Code
claude mcp add --transport http xither https://www.xither.com/mcp
Cursor
Add this to .cursor/mcp.json in a project, or to ~/.cursor/mcp.json for every project:
{
"mcpServers": {
"xither": { "url": "https://www.xither.com/mcp" }
}
}VS Code
Add this to .vscode/mcp.json, or run MCP: Add Server from the Command Palette, choose HTTP, and paste the endpoint URL:
{
"servers": {
"xither": { "type": "http", "url": "https://www.xither.com/mcp" }
}
}ChatGPT
ChatGPT connects to remote MCP servers as custom connectors once developer mode is on in its settings. The menu names have changed more than once, so look for connectors or apps under Settings. Create one with the endpoint URL above and no authentication.
Other clients
Any client that speaks Streamable HTTP can use the endpoint URL directly. A client that only launches local servers can reach it through a bridge such as npx mcp-remote https://www.xither.com/mcp.
What the answers contain
- Every finding is a stored reading: the vendor’s words, the URL of the page and the date Xither read it.
- A page that refused Xither’s reader is reported as refused, with the reason. It is never reported as a page the vendor does not publish.
- A vendor Xither does not watch is reported as not in the record, with a link to the list of vendors it does watch.
- How pages are read, hashed and corrected: methodology.
Watching AI vendors
Know which pages to read and how often.
- Where AI vendors publish their sub-processor lists and DPAs
The URL for each document we read, the date each was last read, and what happened where a page could not be read.
- The AI Vendor Change Lexicon
Sub-processor drift, objection window, silence as consent — each term defined and illustrated with a real clause.
- Which vendor pages to capture, and what each one is worth
Ten page types ranked by measured yield across 1,071 watched pages — including the one we watch 141 times for five date stamps.
- What AI vendors have retired, and where they announced it
235 retirement announcements from 26 vendors, quoted with the page each appeared on — because a retirement is published, not sent.