Remote MCP server connector

The Remote MCP connector is a passthrough: it points at another MCP server and re-exposes that server's tools under your OrgMCP endpoint, namespaced so they can coexist with first-party tools. Its tool list is discovered once at save time (via the remote's tools/list); only a tools/call reaches over the network at runtime.

Connect

In the admin console, open Connectors → Add → Remote MCP server.
Paste the Remote MCP URL (e.g. https://mcp.example.com/mcp). It must be a publicly routable Streamable HTTP (or SSE) endpoint.
Choose an auth scheme: none, bearer, or header (with a header name such as Authorization). Supply the token where applicable.
Click Test connection — a green result reports how many remote tools were discovered — then Add.

Transport

By default the connector uses Streamable HTTP and falls back to SSE if the initial connect fails, mirroring a spec-compliant MCP client. A fresh short-lived connection is opened per call, which keeps the adapter stateless and safe for the serverless request model.

Tool naming & caps

Remote tools are namespaced mcp_<slug>__<remoteToolName>, where <slug> derives from the connector display name (deduped across the server).
At most 60 remote tools are re-exposed (a context-budget guard).
The remote tool's input schema is surfaced verbatim, so clients see the same argument shape the remote expects.

Network & safety

The URL and every resolved DNS record must be publicly routable. Private, loopback, CGNAT, and cloud-metadata addresses are refused (SSRF protection).
Only the endpoint and auth material are stored (encrypted per tenant); the discovered tool list lives in the connector settings.
Remote errors are mapped onto OrgMCP's normalised taxonomy (auth, rate-limit, not-found, unavailable) so failures are explicit rather than silent.

Availability: Team plan and higher. See also Custom API (OpenAPI) for turning a REST API into tools from an OpenAPI spec.