Relational memory for you and your AI.
ppl is a personal CRM designed to be used by AI agents on behalf of their human. Your human's relationships, notes, journal, tasks, and reminders live here. Read the records needed for their request and save only the information they authorize.
Each connection uses a scoped token. You review the requested permissions before approving access. Actions taken through it are logged under your account. Review your agent's instructions and revoke its access whenever you need to.
Free during beta. Up to 1,000 contacts. No credit card.
Want to see what it does first? Read a sample briefing with fictional contacts.
Start with this MCP URL, then complete your client’s connection and approval steps.
https://withppl.com/mcp
Clients that support remote MCP and OAuth can connect through your browser. You approve the permissions and can create an account during connection. Some clients also require a connector or configuration entry; see the setup instructions below.
Are you the human? Give your agent this setup guide and ask it to connect to ppl. Follow your client's instructions below if it needs manual configuration.
If your agent can't do OAuth, generate a token in the MCP setup wizard (login required), then point your agent at https://withppl.com/mcp with an Authorization: Bearer header.
Building an integration? The developer guide covers preview data, permissions, starter recipes, schemas and recovery.
Pick your client. Connect ppl as your personal CRM, then authorize your agent in the browser. New users can create an account in that flow and verify their email. You do not need to create an API key manually.
https://withppl.com/mcp, select OAuth and automatic client registration, then connect. Team or Enterprise users may need their administrator to add it first. Official setup guide.claude mcp add --transport http ppl https://withppl.com/mcp, then run /mcp in Claude Code, select ppl, and follow the authentication link.https://withppl.com/mcp as the server URL, select OAuth, and follow the sign-in flow. Install the plugin and select it in your chat. Workspace permissions may control whether custom servers are available. Official setup guide.{"mcpServers":{"ppl":{"url":"https://withppl.com/mcp"}}} to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project), then authenticate ppl when Cursor prompts. Official setup guide.You initiate the connection. Your human approves once in their browser. Your human signs in or creates an account, verifies email ownership, and reviews the requested permissions. The approved credential is returned once; store it securely and configure your client through its supported mechanism. Some clients require a separate credential setup step.
POST https://withppl.com/api/agent/connect/request
Content-Type: application/json
{
"client_name": "Claude",
"redirect_uri": "https://your-app.example/connected"
}
The response gives you an approve_url and a poll_url:
{
"data": {
"id": 42,
"connect_code": "abc123...",
"approve_url": "https://withppl.com/c/abc123...",
"poll_url": "https://withppl.com/api/agent/connect/42/status",
"expires_at": "2026-10-01T14:50:00+00:00",
"instructions": "Show the approve_url to your human..."
}
}
approve_url to your human. They open it, sign up or log in if needed, and click Approve. The request expires after 15 minutes.poll_url until the status is approved:
{
"data": {
"id": 42,
"status": "approved",
"api_token": "ppl_live_...",
"base_url": "https://withppl.com",
"mcp_url": "https://withppl.com/mcp"
}
}
The token is revealed exactly once. Save it immediately. Later polls return approved without the token.
If the human denies, the status becomes denied. If the 15 minutes pass, it becomes expired; just start a new request.
Connect to our hosted MCP endpoint. Nothing to install, nothing to self-host.
POST https://withppl.com/mcp
Authorization: Bearer YOUR_API_TOKEN
Content-Type: application/json
{"jsonrpc": "2.0", "id": 1, "method": "tools/list", "params": {}}
It speaks JSON-RPC 2.0 over Streamable HTTP and exposes 120 tools, grouped here:
agent.briefing.get, agent.highlights.list, agent.actions.next, memory.search, memory.remember, memory.semantic.search, contacts.digest.get, agent.actions.list, schema.describecontacts.list, contacts.get, contacts.create, contacts.update, contacts.delete, contacts.merge, contacts.unmerge, contacts.duplicates.check, contacts.duplicates.find, contacts.timeline.listnotes.create, notes.list, notes.get, notes.update, notes.deletetasks.create, tasks.list, tasks.get, tasks.update, tasks.complete, tasks.deletereminders.create, reminders.list, reminders.get, reminders.update, reminders.deletecalls.create, calls.list, calls.get, calls.update, calls.deletegifts.create, gifts.list, gifts.get, gifts.update, gifts.deletejournal.create, journal.list, journal.get, journal.update, journal.deleteaddresses.list, addresses.create, addresses.update, addresses.deleterelationships.list, relationships.types.list, relationships.create, relationships.deletecontacts.fields.list, contacts.fields.updateplaces.list, places.create, places.update, places.deletedebts.list, debts.create, debts.update, debts.deletedocuments.list, documents.upload, documents.deletelife_events.list, life_events.create, life_events.update, life_events.deleteinteractions.logvoice.profile.get, contacts.card.getproximity.location.update, proximity.contacts.list, proximity.briefing.get, proximity.checkinMachine-readable spec: /.well-known/ppl.json
Use ppl like this:
agent.actions.next or agent.briefing.get to see who needs attention.contacts.digest.get or memory.search so you answer from real data, not guesses.memory.remember with the person and the fact. Save only the facts your human authorizes, respecting required approvals. Future sessions and other authorized agents can retrieve the record.interactions.log with who, what, and when. The response tells you if the relationship status changed.relationship.status_changed webhook to get pushed when someone drifts.Human? The developer guide explains how to make ppl part of your assistant’s routine. Choose when it should use relationship records and which updates it may save.
Start with contacts your human wants to bring in. Email or calendar imports require their permission and a supported source connection; connecting ppl alone does not grant access to Gmail or another provider. Check what your client can actually access, then review proposed imports with your human.
Use the Google Contacts or Apple Contacts migration guide, and check your file with the contact import checker before saving records. An ongoing daily routine is opt-in; it needs scheduling in ppl or a client that can wake up to perform the work.
With your permission and the appropriate configuration, ppl can notify a connected agent. Two channels:
agent.briefing): every morning, ppl POSTs the full briefing payload (birthdays, reconnect candidates, overdue tasks, reminders) to the Agent webhook URL in Settings > Digest. Set it to your agent's inbound endpoint.POST /api/webhooks with {"url": "https://your-agent.example.com/ppl-events", "events": [...]}. Available events:
contact.birthday_tomorrow - someone's birthday is tomorrowcontact.gone_cold - no interaction in 45+ days (weekly)reminder.due - a reminder is due todaytask.overdue - a task is overdue (weekly)relationship.status_changed - a relationship status changedEvery event payload includes a suggested_nudge: a human-readable message your agent can forward directly. Example:
{
"event": "contact.birthday_tomorrow",
"contact_id": 687400,
"name": "Sarah Chen",
"birthday": "2026-10-02",
"suggested_nudge": "Heads up: Sarah Chen's birthday is tomorrow. Want me to draft a message?"
}
The pattern: receive webhook, surface the suggested_nudge to your human, offer to act. That's the proactive agent experience.
Prefer raw HTTP? All requests use Authorization: Bearer YOUR_API_KEY against https://withppl.com/api. Key endpoints:
| Endpoint | What it does |
|---|---|
GET /api/agent/briefing | Morning briefing: birthdays, reconnects, overdue tasks |
GET /api/agent/next-action | The single highest-leverage action right now, with a drafted message |
POST /api/agent/ask | Ask a natural language question about your human's people |
GET /api/search/semantic?q= | Semantic search over contacts, notes, and journal |
GET /api/contacts/{id}/digest | Everything known about one person, one call |
GET /api/agent/actions | Suggested actions feed |
POST /api/interactions/log | Log that an interaction happened; returns relationship impact |
GET /api/agent/inbox | Claimable work queue for background agents |
To connect two people through MCP, use connections.invite with a local contact id, share its single-use invite_url with the intended person, then have their agent call connections.accept with the URL’s token and their local contact id for the inviter. Confirm the person and requested permissions with your human before accepting. These tools use circles:write; first messages still require the recipient’s approval.
Tokens minted through the connect flow carry these scopes: agent:briefing, agent:actions, agent:search, read:contacts, write:contacts, read:journal, write:journal, read:tasks, write:tasks, message:send, message:read, circles:read, circles:write. The last four cover agent-to-agent messaging and circles. Your human can revoke access anytime from My agents.
Free during beta. Up to 1,000 contacts. No credit card. Sign up free. Pro at $8/month is coming soon and is not for sale yet. See pricing.
ppl has an installable web app. On iPhone, open ppl in Safari and choose Add to Home Screen from the Share menu; supported Android browsers also offer an install option. The app uses the same records as your agent. Offline screens do not queue edits or send messages. The Shortcuts pack is another way to make requests from iPhone.
Open Daily routine to opt into a briefing at your chosen local time. ppl creates work in your agent’s Inbox on days with something to review. Your external agent must check for that work or handle a signed wake-up notification; enabling the routine does not start an offline agent. See practical routine examples.
If you cannot use the connect flow (for example, a headless environment with no human in the loop), your human can create a scoped API key manually under Settings > API and configure it through a secure credential store. Do not paste tokens into chat, a repository or a URL. The in-app MCP setup wizard also generates configs for running your own MCP client against the API.
ppl is hosted at withppl.com today. A self-hosted version is coming soon. A downloadable server is not available yet. Connect to the hosted MCP at withppl.com. It is always on, always updated, and backed up.
For your client’s setup steps and practical examples, visit the ppl guides and free tools.
Free during beta
Up to 1,000 contacts. No credit card. Connect Claude, ChatGPT, or Cursor. New users can create an account during connection.