Privacy Policy — Search Fragments
Search Fragments operates a remote Model Context Protocol (MCP) server at https://searchfragments.com/api/mcp that resolves under-specified or half-remembered queries on behalf of AI agents. This policy explains what data the service handles, how it is used, and who it is shared with.
Who we are
Search Fragments (“we”, “the service”) provides a single read-only tool, resolve_fragment, accessed by software agents over MCP. Contact: hello@searchfragments.com.
What data we collect
- Query content. The text of the fragment submitted for resolution is processed to produce a result and may be retained to operate and improve the service.
- Authentication identifiers. Access tokens are stored only as a one-way hash. We never store raw tokens.
- IP-derived identifier. The caller’s IP address is hashed on receipt. The hash is stored for rate limiting and abuse prevention; the raw IP address is not retained.
- Caller country. The approximate country of the caller is derived from the incoming request using the hosting provider’s geo header and stored with the request log. The raw IP address is not retained; only the country code is stored.
- Request metadata. For each request we log the HTTP method, response status, request path, a request classification, user-agent, referrer, latency, and timestamps. Used for operability, security, abuse prevention, and usage measurement.
- Usage counters. Per-token call counts and downstream API usage (search calls and model tokens) associated with a request.
We do not intentionally collect names, email addresses, or other directly identifying personal information from the calling agent, beyond what an integrator chooses to send in a query.
How we use data
- To resolve queries and return results to the calling agent.
- To operate, secure, debug, and measure the service.
- To enforce rate limits and prevent abuse.
- To improve resolution quality over time.
We do not sell data, and we do not use it for advertising.
Third parties
To resolve a query, the service sends data to:
- Anthropic (Claude API) — classification and semantic processing. Query-derived text is sent.
- Tavily — web search retrieval. The verbatim text of each fragment query is sent to Tavily to retrieve candidate sources.
Infrastructure is provided by Vercel (hosting) and Supabase (database, eu-west-1, Ireland), which process data on our behalf under their own terms. We share data with them only as needed to operate the service. Data stored in Supabase is held in the eu-west-1 (Ireland) region.
Data retention
- Operational logs (request metadata) are retained while the service is operating. We do not currently run an automated deletion schedule.
- Resolved query candidates (the results the service produces) are retained indefinitely as part of the service’s knowledge base, which improves future resolutions.
- Hashed access tokens are retained while an integration remains active, and removed when an integration is closed.
- IP-derived hashes are retained while the service is operating. We do not currently run an automated deletion schedule.
You may request deletion of data associated with your integration by contacting hello@searchfragments.com.
Security
Access to the tool requires authentication. Tokens are stored only as hashes. Traffic is served over HTTPS. Access to underlying data is restricted.
Your choices
Because the service is called by an agent rather than an end user directly, data-subject requests should be made by the integrator operating the agent. To ask what data we hold, or to request deletion, contact hello@searchfragments.com.
Changes
We may update this policy. Material changes will be reflected by the “last updated” date above.