Skip to content

Privacy

Effective date: 14 July 2026. Applies to the Researcher plugin for Claude Code, Claude Cowork, and OpenAI Codex. Maintainer: Marek Sokol, mareksokol98@gmail.com.

Researcher collects nothing. It has no telemetry, no analytics, and no accounts. The maintainer receives no data about you or your work, and cannot: the plugin is a set of skills and local scripts that run on your machine, and nothing in it reports back.

Your manuscript, your bibliography, your data, and your figures stay on your machine. The plugin never uploads them anywhere.

What the plugin sends over the network, and when

Section titled “What the plugin sends over the network, and when”

Outbound requests happen only when a skill or script you invoked needs data to do its job, and only to public scholarly APIs. What is sent is the query itself: a DOI to resolve, a search term, an author name, or an identifier. No identifier of you is attached.

DestinationWhenWhat is sent
api.crossref.orgvalidating a DOI, checking retractions, verifying a referencethe DOI being checked
api.openalex.orgliterature search, verificationthe search query or DOI
api.semanticscholar.orgliterature search, author lookupthe query or author name
export.arxiv.orgliterature search, preprint lookupthe arXiv ID or query
eutils.ncbi.nlm.nih.gov (PubMed)literature searchthe search query

Each is an independent service with its own privacy policy, which governs what it does with a request it receives. The plugin has no agreement with any of them: it uses their public interfaces exactly as a browser would. See the policies for Crossref, OpenAlex, Semantic Scholar, arXiv, and NCBI.

If you never invoke a skill that searches or verifies, the plugin makes no network requests at all.

  • Hooks (active only while the plugin is enabled in Claude Code) read files in your repository and run read-only git commands, to check that your citations resolve. They write nothing and send nothing.
  • Scripts you run yourself write only where you tell them to: install-git-hooks.py writes .git/hooks/pre-commit in your repository, and install-codex-skills.py writes into ~/.agents/ or a repository’s .agents/. Both are reversible with --uninstall.
  • LaTeX compilation runs your local TeX engine on your local files.

The full breakdown of what executes on your machine is in SECURITY.md.

  • It does not collect, store, or transmit your manuscript text, data, figures, or bibliography.
  • It has no telemetry, usage analytics, crash reporting, or anonymous statistics.
  • It does not read, request, or transmit credentials or API keys. External reviewer models (OpenAI, Google, Ollama) are documented as a planned integration and are not implemented, so no keys are read and no requests are made to those services.
  • It sets no cookies and hosts no service. This documentation site is static and served by GitHub Pages, subject to GitHub’s privacy statement.
  • Your AI assistant. Claude Code, Claude Cowork, and OpenAI Codex send your conversation to their providers in order to work. That happens whether or not this plugin is installed, and it is governed by their policies, not this one. See Anthropic’s Privacy Policy and OpenAI’s Privacy Policy.
  • MCP servers you connect yourself. If you connect Scite or a Zotero server, you do so with your own account, and that service’s terms apply to what you send it. The plugin bundles no MCP servers and connects to none on your behalf.

Any change to this policy is committed to the repository, so its history is public and auditable. Material changes are noted in the changelog.

Questions, or a report that something here is inaccurate: mareksokol98@gmail.com, or open an issue.