{
  "markdown": "# Zetesis\n\n[Available on Smithery](https://smithery.ai/servers/reutavidan/zetesis)\n\nScientific due diligence on a claim, from inside Claude, Copilot, or any MCP host.\n\nGive Zetesis a claim, an abstract, a paper, a grant or a deck. It routes the claim to its\nscientific class, then returns the questions a domain reviewer would ask, the failure patterns\nthat caught comparable claims before, and the public evidence bearing on it, with a PMID, DOI,\nNCT number, NIH grant number or SEC filing reference on every source. Every identifier it hands\nback was retrieved. None are generated.\n\nIt can also evaluate a claim **as it stood in an earlier year**, restricting evidence to what\nexisted by then, so a claim is judged on what was knowable at the time rather than on how it\nturned out.\n\n## Connect it\n\nThe hosted server is at `https://api.zetesis.science/mcp`, over Streamable HTTP.\n**No account, key or token is required.**\n\nClaude Code:\n\n```bash\nclaude mcp add --transport http zetesis https://api.zetesis.science/mcp\n```\n\nClaude Desktop (`claude_desktop_config.json`):\n\n```json\n{\n  \"mcpServers\": {\n    \"zetesis\": {\n      \"type\": \"http\",\n      \"url\": \"https://api.zetesis.science/mcp\"\n    }\n  }\n}\n```\n\nAny other MCP client:\n\n| Client | How |\n|---|---|\n| **Microsoft Copilot Studio** | Tools, then Add a tool, then Model Context Protocol. Server URL, auth **None**. |\n| **ChatGPT** | Settings, then Connectors, then Developer mode. Add the URL. |\n| **Gemini CLI** | `gemini mcp add --transport http zetesis https://api.zetesis.science/mcp` |\n\nFor Gemini's `settings.json`, use `httpUrl` rather than `url`; the latter is SSE and will not\nconnect. Full setup notes: <https://api.zetesis.science/docs>\n\n## Tools\n\n**`zetesis_scope`** routes the claim and returns the diligence apparatus for its class: the\nquestions a reviewer would ask, structured by substrate, methods, cohort and risk of bias, a\nfailure-pattern taxonomy carrying the companies each pattern was derived from, and the edge cases\nwhere those patterns were wrong. A checklist that only ever fires positive teaches over-rejection,\nso the counterexamples ship alongside it.\n\n**`zetesis_evidence`** runs the searches and returns a deduplicated bundle from Europe PMC,\nClinicalTrials.gov, openFDA, NIH RePORTER and SEC EDGAR, every source carrying a hard public\nidentifier, followed by the grading rubric so you grade the evidence yourself in context.\n\n**Neither of those calls a language model.** They return in under a second, cost nothing to run,\nand send nothing to a model provider. That is usually the answer a security reviewer is looking\nfor.\n\n**`evaluate_claim`** produces Zetesis's own graded reading server-side. Slower, and only needed\nwhen the assessment itself is the deliverable rather than the evidence.\n\n**`verify_attestation`** re-checks a signed Zetesis record to confirm its claim, evidence and\nconclusion have not been altered since signing. Needs no account.\n\nClaim classes: genomics and Mendelian randomisation, single-cell, bulk omics, CRISPR screens,\nclinical trials, real-world evidence, AI clinical decision support, diagnostics, preclinical\nmodels, cell and gene therapy, structural biology.\n\n## Why the year fence matters\n\nAsk a general model about a 2020 claim today and it answers with years of hindsight; the\npublication that mattered at the time is buried under everything published since.\n\nMeasured on a control claim: unfenced retrieval **missed the pivotal publication entirely** and\nscored 35% evidence coverage. Fenced to the claim's own year, the same query set retrieved it and\ncoverage rose to 79%. So the fence is not only about honesty in retrospect. It is a retrieval\nprecision feature.\n\nSet `as_of` to the year a claim was made for anything that is not brand new.\n\n## Try it\n\n```\nWhat did the published evidence actually support about aducanumab and cognitive\ndecline at the end of 2019, using only sources available by then?\n```\n\nThen ask the same question without the year and compare. The difference is the point.\n\n## Privacy\n\nThe evidence tools send nothing to a model provider. `evaluate_claim` processes claim text through\na model sub-processor, named along with retention terms and hosting region in the\n[privacy policy](https://api.zetesis.science/privacy). Claim text is not logged; only metadata\n(the routed class, depth, counts) is kept.\n\n## The Python client\n\nThis repository also publishes a thin stdio MCP client to PyPI, which predates the hosted server\nand exposes an older tool set (`evaluate_claim`, `check_evaluation`, `verify_attestation`,\n`account_status`). It holds no keys and runs no model; every call is proxied to the hosted engine,\nand it needs a token.\n\n**Prefer the hosted endpoint above.** It needs no token and carries the current tools. The client\nremains for existing stdio setups:\n\n```bash\nclaude mcp add zetesis --env ZETESIS_TOKEN=zk_... -- uvx --from zetesis zetesis-client\n```\n\nTokens: <https://api.zetesis.science/request-access>\n\n- `ZETESIS_TOKEN` sets the token (verification works without one)\n- `ZETESIS_API` overrides the API base, default `https://api.zetesis.science`\n\n## Security\n\nReport vulnerabilities privately to <avidan.r@zetesis.science>. See [SECURITY.md](SECURITY.md).\n\nMIT licensed. The hosted engine is a separate service.\n\n---\nmcp-name: io.github.reutavidan/zetesis\n",
  "bytes": 5326,
  "sha": "b99f02f16f77faae8d3e75910da864e6d12077c382c32677016abeb7d70f1941",
  "repo_slug": "reutavidan/zetesis",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_reutavidan_zetesis_ae7bb7e3/readme"
}