atproto-tailnet
Private mesh networks with AT Protocol identity.
atproto-tailnet is a platform for isolated mesh networks (tailnets). Users sign in with a Bluesky / AT Protocol identity (@handle). There are no passwords and no accounts to manage.
An instance is invite-only. The instance operators get tailnets, and they can let other people create tailnets. Tailnet admins invite people by handle or DID. Other people can request to join the default tailnet of the instance. You can also manage your tailnets from Claude through the remote MCP server.
The code also contains a federation protocol between networks. It is off by default, because it does not carry traffic in the coordinator stack yet.
Quickstart
Start a local instance in less than 10 minutes.
Coordinator Mode
Deploy the Rust mesh coordinator for production.
Connect Claude
Manage your tailnets from Claude with the remote MCP server.
Federation (off by default)
The design and status of links between networks.
Deployment
Deploy to Hetzner, Vultr, DigitalOcean, or AWS with automated orchestration.
Managing Your Network
Invitations, roles, devices, ACLs, security, and instance administration.
Configuration Reference
Every environment variable and config option.
Architecture
How the components fit together.
How it works
-
Sign in with your AT Protocol handle — Your Bluesky identity (
@you.bsky.socialor a custom domain) is your login. You do not need new credentials. -
Get access — An instance operator (
OPERATOR_DIDS) gets a private WireGuard-based mesh network at the first sign-in. A tailnet creator that an operator designates can also create one. Other people see their invitations, or they request to join the default tailnet. -
Invite members — Invite other AT Protocol users by handle or DID. The dashboard gives you an invitation link to share. The service sends nothing by itself.
-
Connect devices — Members generate pre-authentication keys. Then they connect their devices with any Tailscale-compatible client. Each device receives a stable IP in the
100.64.0.0/10CGNAT range. Traffic flows peer-to-peer through encrypted WireGuard tunnels. You can require approval for new devices. -
Manage from Claude (optional) — Add
https://<APP_DOMAIN>/mcpas a custom connector in Claude. Claude signs in as you and has your permissions.
Operating modes
The platform supports two control plane backends:
| Coordinator (recommended) | Legacy Headscale | |
|---|---|---|
| Backend | Single Rust binary | One Docker container per tenant |
| Protocol | Noise IK + HTTP/2 | Headscale gRPC-gateway |
| Multi-tenant | Native, in-process | Container isolation |
| Federation | Off by default (no traffic yet) | Off by default |
| Dependencies | PostgreSQL or SQLite, Caddy | PostgreSQL, Caddy, Docker socket |
| Resource usage | ~16 MB per instance | ~50 MB per tenant container |
Use coordinator mode for new deployments. The project maintains the legacy Headscale mode for existing installations.