Connect an AI client over MCP
Point an MCP client at the public endpoint. It has one tool, ask_octopus, and needs no account.
Try it by hand
The order is initialize, then tools/list, then tools/call. Each is one JSON-RPC request sent with POST.
curl -s https://octo.ad/mcp -H 'content-type: application/json' \
-H 'accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18"}}'
curl -s https://octo.ad/mcp -H 'content-type: application/json' \
-H 'accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'
curl -s https://octo.ad/mcp -H 'content-type: application/json' \
-H 'accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"ask_octopus","arguments":{"message":"What is OCTOPUS?"}}}'
tools/list returns one tool, ask_octopus, with one required input, message, and one optional input, session_id. A tools/call result carries the answer as text in content, and the same answer with follow_up, session_id and status in structuredContent. The fields are listed on MCP tool: ask_octopus.
Steps
- Add a remote MCP server to your client with the address
https://octo.ad/mcp. The transport is streamable HTTP. - Leave authentication empty. The public endpoint needs none.
- List the tools. One tool:
ask_octopus. - Call
ask_octopuswith amessage. A grounded answer from public material, and a follow-up. - Pass back
session_idto continue the same conversation.
What you have at the end
Your client can ask about OCTOPUS. The public endpoint exposes no customer data and no internal graph.
Next
Customer work uses authenticated, tenant-scoped connections, which are not published: Tenant APIs and tools.