About
Why pabloMCP exists
Giving AI agents access to your real accounts without handing any single agent your raw credentials or unlimited authority.
The problem
Agents are genuinely useful the moment they can touch your real accounts — Stripe, GitHub, Cloudflare, Google Workspace, email, SMS, databases. But the usual way to grant that access is pasting long-lived API keys into config files, where every agent gets full authority forever and nothing is logged.
The approach
pabloMCP is a secure capability layer for the Model Context Protocol. Every user gets an isolated workspace with an encrypted vault — AES-256-GCM at rest, per-workspace data keys, so one workspace can never decrypt another's secrets. Agents pair via email + TOTP and receive their own named, revocable bearer keys. Mutating tools preview before they execute, and every write lands in the workspace audit log.
It started as the personal MCP server of Pablo Kraus and grew into a multi-tenant product. It stays deliberately small, transparent about its safety model, and is designed to be read by agents as much as by humans — see llms.txt and the developer docs.
Operator
pabloMCP is built and operated by Pablo Kraus. Reach out via the contact page; how data is handled is on the privacy page.