MCP-GLP-AUDIT-PROXY
AI agents can now read and write in the tools your lab already owns. Without a record of who did what, when, and through which AI tool, that is a data-integrity gap. mcp-glp-audit-proxy is the control layer between the AI brain and your tools: every tool call is checked, optionally approved, and logged in a tamper-evident chain.
AI brain → control layer → tools you already own
The proxy sits between your MCP client (the AI agent) and your MCP servers. Every tool call passes through it.
An allow/denylist decides what may run. Write operations can require human approval (glp_approve). If something is wrong, it fails closed: the call is blocked, not let through.
Each call is appended to a JSONL hash chain (RFC 8785 canonical JSON + SHA-256). The verify CLI detects any tampering, and --print-head lets you anchor the chain head externally.
| Capability | Community (MIT, free) | Pro / Enterprise |
|---|---|---|
| MCP proxy, allow/denylist, fail-closed | Yes | Yes |
| Optional human approval for writes | Yes (glp_approve) | Yes, plus Part 11 e-signatures |
| Audit storage | Tamper-evident JSONL hash chain | + SQL / S3 WORM backends |
| Integrity proofs | SHA-256 chain, verify CLI, --print-head | + Ed25519 signing, RFC 3161 timestamps |
| Reports | RTM + sample OQ report | + PDF/CSV reports, IQ/OQ pack |
| Identity | — | SSO |
| How to get it | PyPI / GitHub | Directly: [email protected] |
--print-head prints the chain head for external anchoring.These materials support your CSV process; your organisation remains responsible for validating its own use.
A short demo, or a pilot on one workflow with the tools you already own.
Request a demo Request a pilotmcp-glp-audit-proxy is designed to support GLP / 21 CFR Part 11 / EU GMP Annex 11 workflows. It does not by itself make a system compliant; validation of each use remains the responsibility of the user organisation.
© Antonio Castellon · [email protected]