Hyperresearch Answers

npx add-mcp 'https://answers.hyperresearch.ai/mcp'

Answers MCP: facts your agent can quote

Answers MCP gives an AI agent two tools at https://answers.hyperresearch.ai/mcp. search runs the open web and about 45 structured sources, from SEC filings to statistics agencies, and returns the passages word for word. ask returns a short answer checked against those quotes, or says it found nothing. Every account gets $5 of free credit, no card needed.

search (5¢ per call)

The full cascade: the open web plus every structured source we connect. Results come back ranked, each with the passages that matched, copied word for word. No summary, no rewriting.

{
  "query": "minimum age to serve as a US senator",
  "max_results": 10
}

Result, shortened:

{
  "results": [
    {
      "rank": 1,
      "url": "https://www.archives.gov/founding-docs/constitution-transcript",
      "title": "The Constitution of the United States: A Transcription",
      "kind": "web",
      "passages": [
        { "text": "No Person shall be a Senator who shall not have attained to the Age of thirty Years", "verified": true }
      ]
    },
    { "rank": 2, "url": "https://constitution.congress.gov/browse/article-1/section-3/", … }
  ],
  "price_cents": 5
}

ask (5¢ to 50¢ per call)

A concise answer with the quotes it rests on. Before it returns, each quote is checked against the page it came from and the answer is checked against the quotes. If nothing supports an answer, you get not_found instead of a guess.

{
  "question": "How old must a US senator be?",
  "mode": "fast"
}

Result, shortened:

{
  "status": "answered",
  "answer": "At least 30 years old (US Constitution, Article I, Section 3).",
  "evidence": [
    {
      "url": "https://www.archives.gov/founding-docs/constitution-transcript",
      "quote": "No Person shall be a Senator who shall not have attained to the Age of thirty Years",
      "check": "verified"
    }
  ],
  "price_cents": 5
}

The open web, then the records behind it

Every search and every ask also queries the structured sources that fit the question. One call, with no source to pick and no extra key to hold.

• Company filings and registers. SEC EDGAR, Companies House, EDINET, HKEX, CNINFO, GLEIF
• Statistics agencies. FRED, Eurostat, e-Stat, World Bank, OECD, IMF, US Treasury Fiscal Data
• Scholarly indexes. OpenAlex, PubMed, Europe PMC, arXiv, bioRxiv, ClinicalTrials.gov
• Books. Open Library, Google Books, HathiTrust, Internet Archive
• Government records. Federal Register, eCFR, GovInfo, Congress.gov, Regulations.gov, USAspending, Grants.gov, openFDA, NPI Registry
• Research funding. NIH RePORTER, NSF Awards, CORDIS, UKRI Gateway to Research
• Code, forums and reference. GitHub, Hacker News, Stack Exchange, X, Wikipedia

The sources a Hyperresearch research run reads: https://hyperresearch.ai/sources

Quotes and an honest not found

• Word-for-word quotes. Every passage is matched against the page it came from. One that does not match is marked unverified.
• An honest not found. When no source supports an answer, ask says so. A not-found answer costs 20% of the call.
• Citable by default. Each answer carries the URL and the exact quote.

{
  "status": "not_found",
  "answer": "No source we could read states this figure.",
  "evidence": [],
  "price_cents": 2
}

Pricing

Call	What it does	Price	If not found
search	Ranked sources with word-for-word passages	5¢	n/a
ask · fast	A quick lookup; returns not found rather than digging further	5¢	1¢–5¢
ask · standard	Digs further on its own when the first pass comes back thin (20¢ when it does)	10¢	2¢–10¢
ask · deep	Longer search; numbers must agree across two sites	50¢	10¢–50¢

• Every account gets $5 of free credit, once. No card needed.
• Top-ups start at $10.
• Auto-reload is optional and stays off until you switch it on.
• A call that fails on our side costs nothing.

Benchmarks

Results coming soon.

Method:

• Harness: the open-source Artificial Analysis Search Index harness (Stirrup), run by us.
• One fixed agent model, the same for every provider, with up to 25 turns per question.
• 30 questions: 10 each from AA-Omniscience (public), DeepSearchQA, BrowseComp.
• The same questions for every search provider; only the search tool changes.

Connect

The server: https://answers.hyperresearch.ai/mcp

Paste this into your coding agent. It adds the server and you sign in on first use:

Fetch and execute the appropriate instructions to set me up for Hyperresearch Answers from https://hyperresearch.ai/answers/setup/prompt.md

The prompt: https://hyperresearch.ai/answers/setup/prompt.md

Or add it directly:

• Add to Claude. Also works in Claude Desktop, and in Claude Code when you are logged in with a claude.ai account.
• Add to Cursor (or the app link cursor://anysphere.cursor-deeplink/mcp/install?name=hyperresearch-answers&config=eyJ1cmwiOiJodHRwczovL2Fuc3dlcnMuaHlwZXJyZXNlYXJjaC5haS9tY3AifQ%3D%3D)
• Add to VS Code

Manual setup

Any client:

npx add-mcp https://answers.hyperresearch.ai/mcp

Claude Code (then run /mcp in a session and choose Authenticate):

claude mcp add --transport http hyperresearch-answers https://answers.hyperresearch.ai/mcp

JSON config, most clients (browser sign-in on first connect):

{
  "mcpServers": {
    "hyperresearch-answers": {
      "url": "https://answers.hyperresearch.ai/mcp"
    }
  }
}

With an API key (create one at https://console.hyperresearch.ai/#/keys):

{
  "mcpServers": {
    "hyperresearch-answers": {
      "url": "https://answers.hyperresearch.ai/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

Windsurf reads serverUrl instead of url:

{
  "mcpServers": {
    "hyperresearch-answers": {
      "serverUrl": "https://answers.hyperresearch.ai/mcp"
    }
  }
}

Answers and Deep

Answers MCP is for the lookups an agent makes many times a task. When the question needs a full report with every citation checked, a Deep research run researches it end to end: https://hyperresearch.ai/pricing

Create an account: https://console.hyperresearch.ai/login