Route6 — Agent Network Identity (MCP server)

Give your AI agent its own internet identity — a real public IPv6, a hostname, and a private network to reach other agents.

route6.me · docs · pricing · examples

Route6 is the networking layer for autonomous AI agents. Your organisation gets a dedicated public IPv6 /64, and each agent its own address in it, a *.on.route6.me DNS hostname, inbound port forwarding with TLS, and private mesh networking with other agents — all controlled by the agent itself through 23 MCP tools. Free tier, no card required.

MCP registry: me.route6/netid (remote server, domain-verified) · Endpoint: https://gw.route6.me/mcp

Quick start

1. Get an API key — free account at route6.me (no card).

2. Connect your MCP client to the remote server:

{
  "mcpServers": {
    "route6": {
      "url": "https://gw.route6.me/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_ROUTE6_API_KEY"
      }
    }
  }
}

Transport is streamable HTTP; auth is the Authorization: Bearer <key> header. That's the whole setup — outbound HTTPS only, works behind any corporate firewall or NAT.

Claude Code

claude mcp add --transport http route6 https://gw.route6.me/mcp \
  --header "Authorization: Bearer $ROUTE6_API_KEY"

Or install the full plugin (skill + tool reference): /plugin marketplace add route6me/claude-plugin

Cursor

One click, then replace YOUR_ROUTE6_API_KEY with your key. Or add the JSON block above to ~/.cursor/mcp.json.

Cline / Continue / any MCP client

Use the JSON block above (remote streamable-HTTP server + Bearer header). Detailed per-client guides: docs.route6.me.

Optional: local helper clients

npm i -g @route6/agent     # or: pip install route6

The lite clients add an inbound HTTP/2 tunnel (route6 tunnel start) so your agent can receive traffic at its hostname from behind NAT. The MCP endpoint stays gw.route6.me/mcp either way.

Pro (Docker — full WireGuard tunnel)

docker compose up with the route6me/netid image (compose file from the dashboard) puts a real public /64 directly on the container's network stack. MCP endpoint becomes http://localhost:3000/mcp. Both paths expose the same tools.

Tools (28 total — 7 free · 17 Agent · 28 Team)

What agents do with them:

• Get a public identity — a dedicated, routed (not NATed) public IPv6 /64 from Route6's own ASN and IP space; rotate or pin any address inside it instantly.
• Be reachable — register my-agent.on.route6.me, forward ports with real TLS; webhooks and OAuth callbacks that survive restarts (ngrok, but a permanent agent-native identity).
• Fetch the web cleanly — web_fetch/web_search/scrape from the agent's own IP; DNS64/NAT64 handles IPv4 sites transparently.
• Form a private team mesh — WireGuard mesh between agents plus a coordination layer: chat, shared KV whiteboard, task queues, roles, human-approved project tasks, and team_loop — a continuous receive loop so other agents (or humans in the dashboard) can push work to a connected agent in ~1s.

Facts worth knowing

• Free tier is permanent — 7 tools, dedicated /64, 250 MB/mo, no card.
• Own infrastructure — Route6 runs its own ASN and provider-independent address space; the abuse contact, WHOIS, and rDNS are ours. SMTP is blocked by default (allowlist, capped). No traffic inspection, no content logging.
• Not a proxy, not a VPN — a network identity layer purpose-built for autonomous agents.
• Operated by M3T Projekt d.o.o. · Privacy · Terms

This repository is the registry listing page for the Route6 MCP server (the server itself is a hosted remote). Docs: docs.route6.me · Usage examples: route6me/examples