Give Claude an evidence layer.

Connect H[1] to Claude and ask questions about scientific research in plain language. H[1] returns structured, traceable answers showing the certainty of the evidence, the studies behind the assessment, and the judgments that produced it.

Nine tools let Claude search and inspect the existing evidence corpus at no additional cost. One tool—graded_research—can evaluate a new research question and uses your research allowance.

Connect H[1] See the ten tools

Connect to the hosted service

Add H[1] as a custom connector in Claude using the MCP endpoint below. You will authorize the connection once through a secure consent screen—no credentials need to be stored in local configuration files.

custom connector URL

https://<your-host>/mcp

Your authorization can be revoked at any time, immediately disabling every connector associated with it.

Run H[1] locally

Developers can run H[1] locally over stdio to explore an existing markdown evidence corpus without authentication or a database. Tools that require the hosted research service are not available in local mode.

terminal

npm install
npm run check     # corpus health
claude mcp add humanofone -- node /abs/path/bin/mcp.js

Ten tools. One connected evidence system.

Nine tools let Claude explore the evidence already assessed by H[1], including certainty ratings, individual studies, risk-of-bias judgments, exclusions, and the methodology itself. These tools are read-only and included on every plan.

The tenth tool, graded_research, commissions an assessment of a new research question. It uses your research allowance only when the question is not already covered by the corpus.

Tool	Answers
`graded_index`	What does this corpus cover, and how certain is each part?
`graded_find`	Which questions here match the one I was asked?
`graded_topic`	One question in full — the rating, every domain judgment behind it, the body, and what was excluded
`graded_search`	Find individual assessed studies
`graded_entry`	One study, with every risk-of-bias domain judgment and its reason
`graded_agreement`	How many qualifying studies point the same way, over an arbitrary slice. Reported separately from certainty and flagged as not part of GRADE.
`graded_ratings`	Every rating in the corpus, one row per question × outcome. Filter by certainty, importance, or the floor
`graded_standard`	The method itself
`graded_research`	Commission a new question. Spends allowance, takes minutes, writes to the corpus
`graded_research_status`	Poll a commissioned job

Never pay twice for the same question. Before starting a new assessment, H[1] checks whether the existing corpus already answers the question. If it does, the result is returned without using your research allowance. As the shared evidence corpus grows, more questions can be answered immediately.

An important safeguard

Every response clearly states that its certainty level describes confidence in the scientific evidence—not the size of an effect, a prediction about a particular person, or medical advice.

H[1] adds this language at the server level, so applications using the system cannot accidentally omit it. Assessments are machine-generated from the published record and have not been independently reviewed by a human unless explicitly stated.