Public documentation in; no customer records out.

Website statistics

The public website uses Vercel Web Analytics to count page views and clicks on tutorial, example and quickstart links. The page URL sent to analytics is the published canonical URL, without query parameters or fragments. Clicks have one of three fixed names: tutorial_open, example_open or quickstart_open. They are signals of interest, not proof that someone installed or used a CRM.

A campaign link can add only two event properties: a source from dev, hashnode, skills, gemini or smithery, and the fixed campaign quote-approval. Other query values, link labels, entered text, customer records and MCP queries are not included in these events. Attribution is not stored in a cookie or local storage and does not persist across pages.

Vercel also processes request metadata to produce statistics about browsers, devices, approximate geography and referral origins. An external referrer with a path, query or fragment causes our script to skip analytics for that page. Vercel describes its visitor hashing and retention in its vercel.com/docs/analytics/privacy-policy. We do not enable analytics cookies, advertising pixels or session replay.

If your browser sends Do Not Track or Global Privacy Control, our script does not load the analytics collector. Blocking JavaScript or the analytics endpoint also leaves navigation and documentation available. Analytics runs only on the public site origin, not on previews, local copies, generated CRM applications or the Docs MCP endpoint. Ordinary hosting requests still reach Vercel when analytics is disabled.

What a Docs MCP request contains

A client sends a JSON-RPC method and, for a tool call, a documentation query, capability identifier or CRM job question. Normal HTTP metadata also reaches the hosting platform, including an IP address, user agent, request time and routing headers. Do not put customer records, credentials, secrets or other personal data into a documentation query: the server neither needs nor asks for them.

What the Docs MCP application code stores

Nothing. The Docs MCP has no database, account, cookie, analytics SDK or write tool. Its application code processes a request in memory and returns public documentation. It does not intentionally persist request bodies, tool arguments, document excerpts or client metadata. Operational logs record only the HTTP method, response status and duration; they do not record the request body or query.

Hosting and retention boundary

The endpoint runs on Vercel. Vercel may process and retain ordinary infrastructure, security and function-log metadata under its own service terms and published privacy policy at vercel.com/legal/privacy-policy. The website analytics described above are separate from the Docs MCP. Accordo does not sell Docs MCP request data or use Docs MCP requests to train a model. Infrastructure retention is controlled by the hosting account and platform, not by the read-only MCP code.

Sharing and subprocessors

The application code calls no third-party API while answering. Request traffic is handled by Vercel as the hosting provider. No CRM provider, model provider, enrichment provider or customer system is contacted by a Docs MCP tool call.

Your controls

You can use the same documentation without the remote MCP through llms.txt, llms-full.txt, the claims ledger and the public repository. Remove the MCP endpoint from your client to stop sending it requests. For a question or deletion request concerning hosting logs, open a private security report through the repository's security policy; the maintainer can determine what the hosting account actually retains rather than promising deletion of data the application never stored.

Changes to this notice

This notice is versioned with the source. A change that adds authentication, persistence, telemetry, a non-public resource or a write tool must update this page and pass review before deployment. Last substantive update: 7 September 2026.