# Search Fragments > A remote MCP server that resolves under-specified, half-remembered, or > fragmentary queries on behalf of AI agents. It saves an agent the search, > read and judge step: one call returns a named answer with a cited excerpt > from the source, a ranked shortlist (ranked_stalls — a valid success), or an > honest decline. Three tools: resolve_fragment, verify_claim, and > submit_resolution_feedback. resolve_fragment typically takes 15–25 seconds. ## What it is Search Fragments is an agent-facing resolution layer, not a human search engine. The consumer is an AI agent; the customer is the developer who operates that agent. ## Tools - resolve_fragment (records each resolution so feedback can refer to it): takes a text fragment, returns one of three honest shapes — Resolved (a named candidate + confidence, with a verbatim excerpt and source link, plus evidence quotes with URLs), ranked_stalls (a ranked shortlist of sources to assemble — a valid success, not an error), or No resolution (an explicit "not resolvable from text"). Every result carries a resolution_id. Expect about 15–25 seconds. Do not treat slowness as failure. - verify_claim: checks a specific factual assertion against current web sources and returns a calibrated verdict with cited evidence and stated limits. - submit_resolution_feedback: attaches recognition feedback (accepted, rejected, or partial) to a prior resolve_fragment result via resolution_id. Stored as feedback only — it does not promote graph facts. ## Report back After the user reacts to a resolve_fragment result, call submit_resolution_feedback once with its resolution_id and accepted (it is what they meant), partial (it helped, not quite), or rejected (not it). Optional: corrected_target, new_clue, note. Send it from the same connection (MCP session) that made the resolve_fragment call. Do not include personal data. Example arguments: {"resolution_id": 1234, "status": "accepted"} ## Endpoint - MCP endpoint: https://searchfragments.com/api/mcp (POST, streamable HTTP) - Access: free and anonymous. Paste the URL and connect. - A GET to the endpoint returns 405 by design. - Registry ID: com.searchfragments/search-fragments ## Connect (anonymous, one paste) - Cursor (~/.cursor/mcp.json): {"mcpServers":{"search-fragments":{"url":"https://searchfragments.com/api/mcp"}}} One click: https://cursor.com/en/install-mcp?name=search-fragments&config=eyJ1cmwiOiJodHRwczovL3NlYXJjaGZyYWdtZW50cy5jb20vYXBpL21jcCJ9 - Claude Code: claude mcp add --transport http search-fragments https://searchfragments.com/api/mcp - Claude Desktop / claude.ai: Settings → Connectors → Add custom connector → https://searchfragments.com/api/mcp (leave the advanced settings empty). Local fallback in claude_desktop_config.json: {"mcpServers":{"search-fragments":{"command":"npx","args":["-y","mcp-remote","https://searchfragments.com/api/mcp"]}}} - Any other MCP-compatible agent: endpoint https://searchfragments.com/api/mcp. Standard url-only config (no account needed): {"mcpServers":{"search-fragments":{"url":"https://searchfragments.com/api/mcp"}}} Clients differ in where the config goes; check your client's MCP settings. ## What people use it for (examples by category) Search Fragments is a free, non-commercial prototype. No account needed. It works on half-remembered songs, films, books, TV, people, historical events, inventions, sport, places, art and culture, and it can also check a specific claim. - All categories: https://searchfragments.com/find - Films: https://searchfragments.com/find/films - Songs: https://searchfragments.com/find/songs - TV: https://searchfragments.com/find/tv - Games: https://searchfragments.com/find/games - Books & poems: https://searchfragments.com/find/books-and-poems - Other languages: https://searchfragments.com/find/foreign-language ## Docs & contact - Docs: https://searchfragments.com/docs - Why agents hallucinate (including SQL / text-to-SQL agents): https://searchfragments.com/agent-hallucinations - Privacy: https://searchfragments.com/privacy - Contact: hello@searchfragments.com ## Pricing Currently free.