human-mcp

A person-centric MCP server. Your context, your rules.

This is a thought experiment, not production software. All data shown here is synthetic. "John Doe" is a fictional persona. This project exists to explore what person-centric context serving could look like — the privacy model, the permission tiers, the audit trail. Do not use this with real personal data. See the README for a full discussion of risks.

The idea

Every LLM conversation starts from zero. Your personal context — who you are, what you're working on, how you communicate — is either absent or locked into a single vendor's custom instructions. A person-centric MCP server puts that context under your control: portable across tools, scoped by permission tiers, and fully auditable.

Person-centric, not tool-centric

Existing MCP servers wrap individual services (one for Gmail, one for Calendar). This wraps a person — one server, all your context.

Permission tiers

Same data, different exposure levels. Full access for trusted tools. Minimal for everything else. You decide.

Audit everything

Every piece of data that leaves the server is logged. You can always answer: "What has any AI learned about me?"

Interactive demo: permission tiers

Select a module and toggle between permission tiers to see how the same underlying data gets filtered.

Module: identity FULL

Simulated audit log

As you browse modules above, each access is logged here — just like the real MCP server would record it.