This page is the human-readable hub. The documents agents actually fetch live
at well-known URLs (and /auth.md at the site root, for the Auth.md
convention). Source for those files is in this repo under agents/.
Start here
- Register. A human creates a workspace at
my.node.uk/signup, then mints a gateway
credential per agent. - Read auth. /auth.md (also mirrored at
/agents/auth.md). - Call the gateway. OpenAI-compatible API at
https://api.node.uk, or the read-only MCP tools at
https://api.node.uk/mcp.
Discovery documents
| URL | What |
|---|---|
| /auth.md | Registration and client_credentials flow |
| /.well-known/oauth-protected-resource | RFC 9728 (this origin; resource = https://node.uk) |
| api.node.uk/.well-known/oauth-protected-resource | RFC 9728 for the gateway |
| /.well-known/api-catalog | RFC 9727 linkset |
| /.well-known/ai-catalog.json | ARD capability manifest (MCP, skill, gateway, agent card) |
| /.well-known/agent-card.json | Discovery card: invoke via MCP, not A2A JSON-RPC |
| /.well-known/mcp/server-card.json | MCP Server Card pointing at api.node.uk/mcp |
| /.well-known/agent-skills/ | Agent Skills index |
| /llms.txt | Site summary for LLMs |
| /ai/ai-gateway/ | Human product page for the gateway |
MCP tools (public, read-only)
Connect a Streamable HTTP MCP client to https://api.node.uk/mcp.
GET /mcp returns the server card and an example (most agents try GET
first). Prefer search over scraping HTML.
search: products and models matching a query (name, vendor, model id)get_product/get_model: one item, with the node.uk page or the
exact OpenAI-compatible URLlist_models/list_products: full public cataloguesget_auth_instructions: pointers to auth.md and OAuth metadata
Billable chat completions stay on the OpenAI-compatible paths under
/api/v1/models/{id}/... with a Bearer token.
The A2A agent card exists so scanners and registries can find us. It
does not speak A2A JSON-RPC: call MCP or the OpenAI-compatible API.