Search Fragments

npx add-mcp 'https://searchfragments.com/api/mcp'

A confident answer, a ranked shortlist, or an honest no.

For queries a model can’t confidently place.

You half-remember something and can’t place it.

Not for: who directed Jaws · capital of France · well-known facts — normal search is faster there.

On the queries a model can’t confidently place, it resolves the answer and points to the sources, narrows it to a ranked shortlist, or declines — rather than filling the gap.

Why AI agents hallucinate

Fragmented queries are a particular hallucination trap. A user gives an agent real signal — a partial memory, an indirect clue, a missing name — but not enough for the model to safely answer from memory. The model can recognise the pattern without being able to verify the answer, and a plausible guess can become a fluent, confidently wrong answer.

Search Fragments is built for that gap: evidence-grounded resolution, a ranked shortlist to confirm by eye, or an honest no-resolution rather than a forced answer. Read: Why AI agents hallucinate and confidently invent answers →

Connect to your agent

Search Fragments is a connector for MCP-compatible agents.

Add this URL to your agent’s MCP settings. When it hits a query it can’t confidently place, it can call Search Fragments instead of guessing.

Free. No sign-up.

https://searchfragments.com/api/mcp

Manual setup

{
  "mcpServers": {
    "search-fragments": {
      "url": "https://searchfragments.com/api/mcp"
    }
  }
}

URL only.

MCP 2025-11-25 Streamable HTTP · Three tools: resolve_fragment · verify_claim · submit_resolution_feedback

Half-remembered, cross-source queries usually take a dozen tabs and five rephrasings. Search Fragments collapses that into one call.

When an agent hits a half-remembered, cross-source query it can’t pin down, it guesses — confidently, and often wrong. Search Fragments is built for exactly those queries. It returns a named answer, ranked sources to confirm by eye, or an honest “not resolvable.” It declines rather than guesses. If your agent has this problem, wire it in and try it — the demo above runs a real resolution right now.

Why Search Fragments exists — one engine, two seekers

• Resolved A named answer, with confidence justified by the evidence found.
• Shortlist Ranked web sources to read and confirm by eye — results returned, but no confident semantic identification made.
• No resolution An explicit no-resolution — not resolvable from the clues given.

Built to decline rather than guess.

Tested

On 50 hard, under-documented fragments, a baseline agent produced three specific fabrications — a real film falsely attributed to a real director who had no connection to it, a specific named sketch that was never performed, and a specific researcher and location for a study that has no single canonical paper. Search Fragments did not assert an answer to any of them.

Building an agent that hits these queries? Get in touch.