← relay-mcp home

Connect an AI assistant

Point your AI client at this relay's MCP endpoint and it can use the tools on every machine you've linked. The endpoint is:

https://mcp.darky.cc/mcp
You need an access token. Ask your relay administrator for one — they create it in the admin console and choose exactly what it may do on each machine. In the examples below, replace <TOKEN> with the token you were given. Give each client its own token. Transport is Streamable HTTP.

Command-line & config-file clients

Claude Code direct

Anthropic's CLI (and the Claude Code built into the Claude desktop app).
claude mcp add --transport http relay https://mcp.darky.cc/mcp \
  --header "Authorization: Bearer <TOKEN>"
Add --scope user to use it in every project. Verify with /mcp in a session. Tools appear like opencode1__diagnostics_system.

OpenAI Codex CLI direct

Edit ~/.codex/config.toml:
[mcp_servers.relay]
url = "https://mcp.darky.cc/mcp"
bearer_token_env_var = "RELAY_TOKEN"
…then export RELAY_TOKEN=<TOKEN>. Or inline a static header instead:
[mcp_servers.relay]
url = "https://mcp.darky.cc/mcp"
http_headers = { "Authorization" = "Bearer <TOKEN>" }

OpenCode direct

Add to opencode.json (project root) or ~/.config/opencode/opencode.json:
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "relay": {
      "type": "remote",
      "url": "https://mcp.darky.cc/mcp",
      "enabled": true,
      "headers": { "Authorization": "Bearer <TOKEN>" }
    }
  }
}
Tip: use "Bearer {env:RELAY_TOKEN}" and export RELAY_TOKEN to avoid hardcoding.

OpenClaw direct

openclaw mcp add relay \
  --url https://mcp.darky.cc/mcp \
  --transport streamable-http \
  --header "Authorization:Bearer <TOKEN>"
Check with openclaw mcp doctor relay --probe. Leave OAuth off when using a static token.

Hermes (Nous Research) direct

Add to ~/.hermes/config.yaml:
mcp_servers:
  relay:
    url: "https://mcp.darky.cc/mcp"
    headers:
      Authorization: "Bearer <TOKEN>"
Reload in-session with /reload-mcp, or hermes mcp test relay. Requires Hermes v0.2.0+.

Desktop & web apps

Claude desktop app (Cowork + Code) direct via Claude Code

The graphical Custom Connectors dialog is OAuth-only, so use the bundled Claude Code for a static token — open its terminal and run the Claude Code command above. (Or configure it in ~/.claude.json / %USERPROFILE%\.claude.json with a "headers" block.)

Claude.ai (web) beta

Settings → Connectors → Add custom connector:
  1. Remote MCP server URL: https://mcp.darky.cc/mcp
  2. Under Request headers (Advanced): add Authorization = Bearer <TOKEN> — include the literal word Bearer.
  3. Enable it in a chat via + → Connectors.
Static-header auth is in beta — if only OAuth fields appear, your account doesn't have it yet. Requires a paid plan; the server is reached from Anthropic's cloud (this endpoint is public, so that's fine).

ChatGPT (web) workaround

ChatGPT's connector UI (Settings → Apps & Connectors → Developer mode → Create) only offers OAuth or No authentication — there is no static-token field. Two options:
  1. Front this relay with a small proxy that injects Authorization: Bearer <TOKEN>, then register the proxy's URL with No authentication; or
  2. Put an OAuth 2.1 gateway in front and register that.
Requires the beta Developer mode and a paid plan. Ask your admin — they may already expose an OAuth-fronted URL for ChatGPT.

Once connected

Ask naturally, e.g. "using the relay, show diagnostics for opencode1". What each token can actually do on each machine is set by your administrator. If tools don't appear, your token may not be granted access to any client yet — ask your admin.