Open MCP Portaldocs

Building gateways

Create a team gateway in Open MCP Portal, choose its capabilities from your servers and the catalog, set minimum roles, rate limits and recording, and connect to it.

A gateway is a curated set of capabilities at one MCP endpoint, <portal URL>/mcp/{slug}. Team owners and admins create and edit the team's gateways under Gateways; every member of the team can read them and connect. Platform admins build platform gateways the same way and grant them to other teams, people or SSO groups.

Team gateways are for the team's own members only. To share tools with other teams, publish the server and ask a platform admin for a platform gateway.

Create a gateway

Gateways › Create gateway:

  • Name and Description (people see them under My gateways).
  • Slug: 1 to 48 lowercase letters, digits and hyphens, unique across the platform. It can't be changed later: the endpoint is /mcp/<slug>.
  • Enabled: turned off, the endpoint answers "not found" to everyone.

Then choose its capabilities.

Choose capabilities

The Capabilities tab's Choose capabilities picker offers:

  • any capability of your team's own servers, private or published, as long as the server is enabled and still lists it; and
  • approved capabilities of published catalog servers, from any team.

Search by name, title or server, filter by Server and Kind, and use Selected only to review what you picked. Badges show what each capability needs: Needs GitHub (repo) for a user authorization requirement, a sign-in for OAuth servers, Server records contents when its server records call contents. Your selection is kept while you search and page.

Rules when you save:

  • Names must be unique within the gateway, ignoring case and across kinds. Clients see tools and prompts as {server-slug}__{name} and resources as omp://{server-slug}/{uri}, so two servers never clash, but two capabilities of one server with names differing only in case do.
  • Resource templates are offered only when they use simple {var} or reserved {+var} expressions with one variable each.
  • A capability already selected can stay even while it's unavailable (for example its server left the catalog): every call checks it again, and it comes back when it's available.
  • If someone else saved the gateway while you were editing, the picker says so; saving puts yours over theirs.

Untrusted text

Names, descriptions and schemas come from the servers, and a model reads them as instructions. Read them before you add a capability.

Minimum roles

On a team gateway, each capability has Who may use: Every member (the default), Editors and above, Admins and owners or Owners only. Callers below the minimum don't see the capability, and their calls are refused. A team service account counts with its own role (member or editor). Platform gateways have no minimum roles: their grants decide who may use them.

Availability

A selected capability can be unavailable for a while; the Capabilities tab says why, and calls to it are refused:

Shown asMeans
No longer listedThe server doesn't list it any more
Team archivedThe server's team (or this gateway's team) is archived
Server disabledIts server is turned off
Left the catalogA catalog capability whose server was taken off the catalog
Not approvedA catalog capability whose current definition isn't approved
Changed, waiting for approvalIts definition changed since it was approved; it's served again once the change is approved

Capabilities of your team's own servers need no approval in your team's gateways.

The gateway's tabs

  • Overview: the endpoint URL, the gateway at a glance (capabilities, unavailable now, status), whether it records call contents, and a Connect card with snippets. See Connect your AI client.
  • Capabilities: the selection and its availability.
  • Activity: the gateway's calls (owners and admins). See Activity and usage.
  • Settings: below.

Settings

  • General: name, description and Enabled. Turned off, the endpoint answers "not found" to everyone; its capabilities and access stay as they are.
  • Rate limit: requests per minute from all callers together, 1 to 1,000,000. Inherit the team's or platform's default, or set This gateway's own limit, up to the platform's ceiling. The form shows the value that applies now and after saving. Over it, callers get HTTP 429 and are told to retry shortly; each person, key and service account also has its own limit. See Limits.
  • Record call contents: off by default. See Recording call contents.
  • Delete this gateway: removes it and its endpoint. Clients connected to it stop working at once, and its slug becomes free. Its call records are kept.

Platform admins can turn a team gateway off and on from Admin › Gateways, and nothing else.

Who can use it, checked every time

Every request to a gateway checks, in order: the credential, the person (or service account), that the gateway is on and its team active, that the caller may use it, that the capability is still selected and available, the caller's role, and the rate limits. A refused call looks like a call to a tool that doesn't exist. Changes take effect at the next request, with no reconnect.

On this page