Documentation

Zetesis performs due diligence on a high-dimensional scientific claim, graded dimension by dimension against the published record. Give it a claim, a paper, an abstract or a pitch deck. It works out what would have to be true for the claim to hold, then reports what the evidence shows on each of those dimensions, what it does not show, and the ways claims like it have failed before. Every identifier it returns was retrieved from a public register. None are generated.

Connect it

A remote MCP server over Streamable HTTP. No account, key or token is required.

https://api.zetesis.science/mcp

Claude Code

claude mcp add --transport http zetesis https://api.zetesis.science/mcp

Claude Desktop, in claude_desktop_config.json:

{
  "mcpServers": {
    "zetesis": { "type": "http", "url": "https://api.zetesis.science/mcp" }
  }
}

Client	How
Microsoft Copilot Studio	Tools, then Add a tool, then Model Context Protocol. Server URL, authentication None.
ChatGPT	Settings, then Connectors, then Developer mode. Add the URL.
Gemini CLI	gemini mcp add --transport http zetesis https://api.zetesis.science/mcp
Grok	grok.com/connectors, then New Connector, then Custom. Paste the URL, authentication None.

For Gemini's settings.json, use httpUrl rather than url; the latter is SSE and will not connect.

Example prompts

Ask in plain language. The assistant decides when to reach for Zetesis; you do not call the tools by name.

1. Judge a claim on what was knowable at the time. This is the one worth trying first.

What did the published evidence actually support about aducanumab and cognitive decline at the end of 2019, using only sources available by then?

Then ask the same question without the year and compare. Unfenced, the report that mattered at the time competes with everything published since and gets buried.

2. Check a claim before repeating it.

A colleague told me intermittent fasting improves insulin sensitivity independent of weight loss. Is that actually well supported?

3. Diligence on an asset or a deck.

Here is a licensing sheet for a cardiac regeneration gene therapy. What would have to be true for the central claim to hold, and what does the published record actually show?

4. Check a paper before citing it.

Before I cite this trial, has it been retracted or corrected, and did the result replicate?

The tools

Tool	What it does
zetesis_scope	Routes the claim to its scientific class and returns the diligence apparatus for it: the questions a domain reviewer would ask, the failure patterns that have caught comparable claims before, and the edge cases where those patterns were wrong. Calls no language model.
zetesis_evidence	Runs the searches and returns a deduplicated bundle with a hard identifier on every source, followed by the grading rubric. Calls no language model.
evaluate_claim	Produces Zetesis's own graded reading server side, for when the assessment itself is the deliverable rather than the evidence. Slower, and requires a token.
verify_attestation	Re-checks a signed Zetesis record to confirm its claim, evidence and conclusion have not been altered since signing.

All four are read-only. None writes to any system or takes a destructive action.

The first two are the recommended route and are not metered. They return in about a second, and the assistant reads the sources at full depth in its own context.

What it covers

Genomics and Mendelian randomisation, single-cell, bulk omics, CRISPR screens, clinical trials, real-world evidence, AI clinical decision support, diagnostics, preclinical models, cell and gene therapy, and structural biology.

Sources: Europe PMC, ClinicalTrials.gov, openFDA, NIH RePORTER and SEC EDGAR. Every source is checked against retraction and correction notices, and the notice is cited when one exists.

Authentication

The endpoint above needs none. An authenticated endpoint also exists at https://api.zetesis.science/mcp-secure, which requires OAuth 2.1 and supports dynamic client registration; clients that handle OAuth will discover it from the WWW-Authenticate header. A token for evaluate_claim can be requested at /request-access.

Data handling

• zetesis_scope and zetesis_evidence call no language model and transmit nothing to any model provider. Claim text is turned into search terms for public registers and the results are returned.
• evaluate_claim processes claim text through a model sub-processor, named along with retention terms and hosting region in the privacy policy.
• Claim text is never written to our usage records. What is recorded: timestamp, which tool was called, the routed claim class, your IP address and the client name your software sends. These are used to rate-limit the free tier and to keep the service running.
• Nothing is stored on your system. Outbound HTTPS to api.zetesis.science only.

Limits

• Five public registers and nothing else. A claim resting on unpublished or proprietary data returns as unshown, which is not the same as disproven.
• Abstract level. It reads what the registers return, not full text.
• Retrieval can be degraded when an upstream register is unavailable; the response says so, and the reading should then be treated as partial coverage.

More

Privacy policy · Terms of use · HTTP API reference · Source

Questions: avidan.r@zetesis.science