Other MCP clients

Beacon's MCP server speaks streamable HTTP at /mcp, stateless, so any client that speaks MCP over HTTP reaches it. In VS Code, for example, put this in .vscode/mcp.json:

{
  "servers": {
    "beacon": {
      "type": "http",
      "url": "https://status.example.com/mcp",
      "headers": { "Authorization": "Bearer lbt_…" }
    }
  }
}

How a client signs in

There are two ways, and a client needs either one of them:

• MCP's OAuth, for a client that implements it (Claude Code, Codex and Cursor do; see their pages). Beacon publishes the protected-resource metadata of /mcp and the authorization server's metadata (RFC 9728, RFC 8414), registers clients dynamically as public clients (RFC 7591) with redirect URIs on the loopback interface or https, and takes the authorization-code flow with PKCE (S256). The consent page asks what the agent may do — the same four things as for Claude Code (Allow it on Beacon's consent page).
• A token in the header, Authorization: Bearer lbt_…, for everything else: a script, a CI job, a client without OAuth. The token is one of your agents' credentials, and it renews itself (Headless agents and CI). An OAuth access token (lba_…) is accepted wherever a bearer token is, so a client that has signed in can also use it as a header.

Check it works

curl -H "Authorization: Bearer lbt_…" https://status.example.com/api/v1/token

answers who the token acts for, as its holder sees it; the MCP tool whoami says the same. Every tool takes a scope, checked on every call, and every call is recorded: MCP tools lists them all with what they need.