Custom tools and prompts
Beyond one tool per endpoint, MCP Studio lets you publish tools, prompts and resources you design yourself. Use them to give agents a whole task as one step, instead of leaving them to work out the order of several calls.
All three are drafts until you publish them, like everything else on the page.
Chained tools
A chained tool runs several endpoints in order as one tool. Use one when a job always takes the same calls in the same order, such as creating a customer and then invoicing them.
| Field | What it does |
|---|---|
| Tool name | What agents call it. Letters, digits, _ and - |
| Description | What it does, and when an agent should pick it over the single calls |
| Arguments | What the agent passes in, each with a name, a description and whether it's required |
| Steps | The endpoints to call, in the order they run |
| Returns | What the agent gets back |
| Tags | Comma separated. Tool groups take everything carrying one of their tags |
Passing values between steps
Use Insert reference to fill a field from an argument or an earlier step:
{{args.email}}
{{steps.customer.body.id}}
{{steps.list.body.items[0].id}}Approval
The person approves the whole chain once, with every request listed. Values from later steps aren't known yet, so they're shown as references such as {steps.customer.body.id}.
A chained tool obeys the same permissions and limits as any other, and it runs inside any group.
Prompts
A prompt is a ready-made conversation starter. Most clients show prompts as slash commands.
| Field | What it does |
|---|---|
| Prompt name | What a client lists and asks for |
| Title | Its readable name, such as "Create a customer and invoice them" |
| Arguments | Values to fill in when it's used |
| Messages | The text, each message from User or Assistant |
Use arguments inside a message like this:
Create a customer for {{args.email}}, then invoice them.Resources
A resource is reference text an agent can keep open. It can serve the OpenAPI spec or a guide you've published. Add a description saying when an agent should read it.
Generating from your guides
If you've already written the steps down in a guide, you don't have to rebuild them by hand. Guide workflows makes the chain and prompt for you.
Testing
What agents see shows how each item will appear. To check that agents really choose and use it, add it to a test run.