Connect a client

Every connector is a remote MCP server over HTTP. Any MCP-compatible client can call it — there is no SDK to adopt and no bespoke protocol to learn.

Pick a client, then a set of connectors

Go to Agent clients, choose the client your team uses, and tick the connectors you want. You get a config block to paste. The same page tells you which auth mode that client supports.

Clients differ in what they accept — some take remote MCP directly, some need a local bridge, some support OAuth and some only API keys. The list is checked per client rather than assumed, and a client is only marked supported once its remote-MCP and auth path has actually been verified.

Authentication

Connectors sit behind OAuth 2.1 with PKCE and dynamic client registration. For clients that support it, this is the whole setup: the client registers itself, you approve once in the browser, and tokens refresh on their own.

For clients that cannot do OAuth, use an API key from Access & limits.

What a config looks like

The exact shape is generated per client on the harness page — copy it from there rather than hand-writing it, because the field names differ between clients. Broadly:

{
  "mcpServers": {
    "mospi": {
      "url": "https://platform.rsi-data.com/marketplace/mospi/mcp",
      "transport": "http"
    }
  }
}

Trying one before you wire it up

Every connector has a playground — open any card from Discover and use Try it. It runs the same tools your agent would call, so a question that works there works from your client.

The first request to a connector may take a few seconds: the zones scale to zero, and a cold one has to wake. Subsequent calls are warm.

Which connectors should I connect?

Fewer than you think. Every connected server adds tools to your agent's context, and an agent given 393 tools chooses worse than one given 20.

Start from the question rather than the catalogue: ask the consultant what answers it, and connect what it names.