people-context: memory about the people in your life

Ask an assistant how to raise a reporting delay with Priya and it has nothing to go on. It doesn’t know which Priya, that she’s your contact at a partner org, that you agreed a new deadline last week, or that she prefers a short email to a call. That knowledge is spread across your head, your inbox and a notes file. people-context stores only this kind of memory, about people, behind a few narrow tools, in one SQLite file on your machine.

What it does

• Who is this: explainable resolution over names, nicknames, aliases and handles. Two Priyas come back as two candidates, each with the reason it matched, rather than one guess.
• What you know: relationships, organizations and roles, durable facts, interaction summaries, traits, reminders and a per-person timeline, revealed in layers as the request requires.
• How to approach them: communication guidance based on recorded traits, past friction and open follow-ups, plus reports on relationships gone quiet and upcoming dates.
• Safe imports: email, vCard, calendar, LinkedIn, Outlook and WhatsApp exports become candidates you review; nothing is recorded until you approve, and raw source content isn’t kept.

Who it’s for

• Salespeople, consultants and recruiters who manage many client, partner or candidate relationships.
• Managers who want a reminder of where things stood before a meeting.

Setup

Needs Python 3.11+ and uv. Generic config for Cursor, Windsurf, VS Code and similar clients:

{
  "mcpServers": {
    "people-context": {
      "command": "uvx",
      "args": ["--from", "people-context", "people-context"]
    }
  }
}

Claude Code and Codex can install it from a plugin marketplace. To try it first, uvx --from people-context pctx demo --reset seeds a fictional dataset.

Our take

Only a few dozen stars as of 2026-10-06, and created in July 2026, but it’s one of the clearest-thinking memory projects in this batch. It skips general memory, sticks to relationships, and treats trust as a design problem: every write leaves an audit trail, forget actually deletes, sensitive records need an operator unlock that no prompt can trigger, and everyday commands stay offline. Against the official Memory server’s generic graph, the name disambiguation and reviewed imports are real differences. Keep in mind that these are your notes about other people, so think before importing contacts and chat logs. The project is young and its interfaces may change. Licensed MIT.