Other MCP clients

Any client that speaks MCP's Streamable HTTP transport can connect. This page has what you need to configure one that is not in the Studio panel.

Connection details

Setting	Value
Transport	Streamable HTTP (stateless — no session id)
Endpoint	https://mcp.hypervalent.io/mcp?org=<org id>
Authentication	Authorization: Bearer atk_… or X-API-Key: atk_…
Organization	?org= on the URL, or an X-Hypervalent-Org: <org id> header
Protocol versions	The current MCP specification and the previous revisions the SDK supports

The org id is the organization's public id shown in Studio's Use with AI panel and on the API tokens page — a short number, not a UUID.

Every request is independent: the server keeps no session between calls, so a client may reconnect at any time and load-balanced instances behave identically.

Stdio-only clients

A client that can only launch local (stdio) servers can use the mcp-remote bridge, exactly as Claude Desktop does:

npx -y mcp-remote "https://mcp.hypervalent.io/mcp?org=YOUR_ORG_ID" \
  --header "Authorization: Bearer atk_YOUR_TOKEN"

Testing with curl

A raw initialize call is a quick way to confirm the token and org are accepted before touching a client:

curl -s https://mcp.hypervalent.io/mcp?org=YOUR_ORG_ID \
  -H "Authorization: Bearer atk_YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"curl","version":"0"}}}'

A good answer is a JSON-RPC result with the server's capabilities. A bad token answers a JSON-RPC error whose data carries the API's { status, code, message }.

GET https://mcp.hypervalent.io/health answers {"ok":true} and needs no token.

Building your own client or agent

• The tool catalog is fixed per token: tools/list always returns every tool, and the server refuses the ones the token cannot use at call time (a read-only token gets TOKEN_SCOPE_READ_ONLY on writes). Filter on the annotations if you want to hide writes.
• Tool results are { content: [{ type: "text", text: "<JSON>" }] }; failures set isError: true and the text is the error envelope described in Errors & limits.
• Resource reads that fail are JSON-RPC errors — -32002 for a missing resource.
• The server rate-limits POST /mcp per token as a backstop, separately from the API's own limits; both answer a Retry-After. See Errors & limits.