Does the Peerscy MCP server run in the cloud?

No. It is a local process your MCP host launches over stdio. Nothing is hosted, and no manuscript data leaves your machine.

How do I add it to Claude Code or Cursor?

Install Peerscy, then configure the host to run 'peerscy mcp'. In Claude Code: claude mcp add peerscy -- peerscy mcp.

Peerscy MCP server

A local Model Context Protocol server for scientific peer review.

Peerscy ships a Model Context Protocol server on stdio. MCP hosts such as Claude Code and Cursor launch it as a local process and call Peerscy's deterministic, model-free tools. There is no hosted service: nothing to deploy, no domain, and no manuscript data leaves your machine.

What it is

Peerscy is a private, local-first AI peer-review assistant. Its MCP server exposes the same verification tools the desktop app uses, so an agent can check statistics, references, and the literature without sending your work to a cloud. It runs in-process via llama.cpp; the exposed tools do not need a loaded model.

Tools

Tool	What it does	Network
`statcheck`	Recompute p-values from reported t/F/χ²/r/z statistics and flag inconsistencies.	Offline
`calculator`	Evaluate a mathematical expression exactly.	Offline
`multiple_comparisons`	Bonferroni / Holm / Benjamini–Hochberg correction of a p-value family.	Offline
`effect_size`	Convert between r, Cohen’s d, f², η², and odds ratio.	Offline
`power`	Power, or required sample size, for t-tests and correlations.	Offline
`paper_info`	Metadata, citation count, and retraction status from a DOI or paper link.	Keyless API
`search_literature`	Search OpenAlex, Europe PMC, arXiv, PubMed, Semantic Scholar.	Keyless API
`read_url`	Read a URL; paper links resolve to their citation in one call.	Keyless API
`check_citations`	Extract DOIs from a reference list and check each for retraction or validity.	Keyless API

The deterministic tools (statcheck, calculator, multiple_comparisons, effect_size, power) never touch the network. The others call keyless public APIs with the arguments you pass.

Run it

Install Peerscy, then run the server. It is headless and speaks newline-delimited JSON-RPC on stdin/stdout.

peerscy mcp

Connect a client

Claude Code:

claude mcp add peerscy -- peerscy mcp

Any MCP client (JSON config):

{
  "mcpServers": {
    "peerscy": {
      "command": "peerscy",
      "args": ["mcp"]
    }
  }
}

Point command at the Peerscy binary (the packaged app, or cargo run -- mcp from a checkout).

Privacy

The MCP host talks to a local Peerscy process. Manuscripts and prompts are not sent to any server operated by the Peerscy project. Only the tools marked “Keyless API” make outbound requests, and only with the arguments you pass. See Security & Trust for the full list of network and storage paths.

FAQ

Does it run in the cloud? No. It is a local stdio process; nothing is hosted.

Do I need an API key? No. The tools are keyless.

Which tools does it expose? statcheck, check_citations, paper_info, search_literature, read_url, calculator, multiple_comparisons, effect_size, and power.

Download Peerscy