Skip to main content
Feedback

Creating and managing Skills

The Skills page in the MCP Gateway user interface is the workbench to do the following actions:

  • Create: A guided form for name, description, body, and the capability group.
  • Import: Upload an existing SKILL.md file. MCP Gateway parses its YAML frontmatter (name, description) and body. You attach a capability group afterward.
  • Edit: Updates instructions and the capability group independently (PUT /:id/details versus PUT /:id/capabilities at the API level).
  • Delete: Removes a Skill you own.

Drafts

Editing a saved Skill creates a draft overlay instead of changing it directly, so in-progress edits do not affect what is currently served. Saving a draft uses optimistic locking against the Skill's last known updatedAt value. If someone else edited the Skill in the meantime, MCP Gateway rejects the save as a conflict instead of silently overwriting the change.

Scoping: Who a Skill is served to?

Skills use MCP Gateway's existing consumer-permission model, the same one used for tool access, instead of an organizational hierarchy.

You can enable a Skill for one of the following:

  • consumerTag: A named group of consumers you are already tagged in MCP Gateway. For example, devs and ops.
  • clientName: The specific MCP client application to which you are connecting. For example, cursor and claude-desktop.

Enabling a Skill for one tag or client name does not expose it to others. A Skill with no matching enablement for the current consumer is invisible to it. For such Skills, neither the resource nor the prompt is served.

On this Page