The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Graneth listing page.
Source for @graneth/mcp-server — an account-free MCP
server that checks AI-written changes before they are committed: packages that do not
exist, dependency risk shapes, and hardcoded secrets.
Tool documentation, install snippets for Claude Code / Cursor / Windsurf, and the exact
verdicts it returns: packages/mcp-server/README.md.
packages/mcp-server | the MCP server itself — stdio, hand-rolled JSON-RPC, no runtime dependencies |
packages/core-checks | the detection core it bundles: registry existence across six ecosystems, manifest parsing, secret patterns + entropy, risk scoring |
The hosted scanning service is a separate, closed codebase: pull-request scanning, the second-pass LLM triage, policy enforcement, signed receipts and billing are not here and are not open source. This repository is the free tool and the checks it runs — which is the part you install on your machine, and therefore the part worth being able to read.
Generated from the internal monorepo, which is the source of truth. Edits made directly here are overwritten on the next sync, so please open an issue rather than a patch — the fix lands upstream and comes back with the next release.
Three of core-checks' test files are held back: they assert that the secret rules fire,
so their fixtures are literal example credentials, and every scanner — GitHub's push
protection and this project's own — reads them as the real thing. The rules themselves
are here in full (packages/core-checks/src/secrets.ts), as are the package's other tests.
MIT. See LICENSE.