Skills and MCP tools
Skills and MCP tools in the Prompt Verse library: a description and inputs for a skill, and a mock or an HTTP call for a tool you author.
A skill and an MCP tool are two of the five kinds of asset in your library, next to prompts, agents and workflows. They are not the same thing as the tools your AI sees when it connects to Prompt Verse.
Skills#
A skill is know-how an agent can be handed as a tool. You give it a title. Its data holds:
- a description of what it does
- an optional kind:
function(the default),hosted,mcporagent - an input schema: JSON Schema as text, or a short sketch of the arguments in plain words
The starter Look up an order is a skill. A skill compiles to a tool definition. It does not run a model by itself. In the pre-beta your own AI does the running: it reads the skill and uses that definition.
MCP tools you author#
An MCP tool is a contract you write: a name, a description and an input schema, plus how a call is answered.
| Execution | What a call does |
|---|---|
mock | Returns the response you stored. Use this before an endpoint exists. The starter Order status is a mock. |
http | Sends the input to the endpoint you saved. An HTTP tool without an endpoint is refused. |
A failed call is reported. Prompt Verse does not invent a successful answer when the call does not succeed.
Shelves, not a second library#
Skills and MCP tools take tags, and a collection can hold them (kind is skill or mcpTool). Taking one off a shelf does not delete it. Plan counts cover prompts, workflows and agents. See Plans and limits.
For your AI#
The library tool lists and creates them: skills, skill, create_skill, tools, create_tool and call_tool. call_tool takes the tool’s name and an optional input. See the library reference.
Over HTTP the same nouns are GET /api/skills, POST /api/skills, GET /api/tools and POST /api/tools/:name/call. See The HTTP API.