# AI and MCP Tool Designer

> Design, run, and test AI and MCP tools in your browser.

AI and MCP Tool Designer is a browser-based editor for defining AI-callable and MCP tool contracts with JSON Schema, implementing them with JavaScript or static responses, and testing tool calls in an AI conversation; it does not start an MCP server.

Also: ai and mcp tool designer, ai mcp tool designer, ai tool designer, mcp tool designer, ai tool builder, mcp tool builder, tool calling playground, function calling playground, json schema tool builder, ai agent tool testing, browser tool sandbox.

## What it does

AI and MCP Tool Designer puts the contract, implementation, and AI testing workflow for AI and MCP tools in one workbench. A tool needs a callable name, an accurate model-facing description, a JSON Schema input contract, and an implementation that returns a useful result. Tool definitions are saved in IndexedDB on this browser, JavaScript runs in Postman Sandbox, and anonymous AI testing uses a limited number of free turns through the Postman gateway. The browser tool designs and tests MCP-compatible tools but does not expose them through a local MCP server; Postman adds that capability.

## When to use it

- You are prototyping tools for an AI agent and want to test their schemas before integrating them.
- You are designing tools for an MCP server and want to validate each tool contract and implementation first.
- You want to compare a scripted tool with a deterministic static-response tool.
- You need to see the exact arguments an AI generated and the result returned to it.
- You want a local draft that survives refreshes without creating an account.

## What you give it

| Input | Meaning |
| --- | --- |
| AI or MCP tool definition | A callable name, model-facing description, and JSON Schema object describing the accepted arguments. |
| Tool implementation | JavaScript executed in Postman Sandbox or a deterministic static JSON response. |
| Playground prompt | A natural-language request sent to the anonymous Postman AI gateway with the selected tool definitions. |

## What you get back

| Output | Meaning |
| --- | --- |
| Validated tool result | The local result or validation error produced for supplied sample arguments. |
| AI tool-call trace | The generated tool arguments, execution status, and result used by the assistant. |
| Browser-local toolset | A saved IndexedDB draft restored when the tool is reopened in the same browser. |

## Key terms

- **Tool calling** — a model choosing a named function, generating arguments that satisfy its schema, receiving the function result, and continuing its response with that result.
- **Model Context Protocol (MCP)** — an open protocol that lets AI applications discover and call tools exposed by an MCP server using structured names, descriptions, input schemas, and results.
- **JSON Schema** — a machine-readable description of the object a tool accepts, including property types, required fields, enums, formats, and constraints.
- **Static-response tool** — a deterministic tool that always returns configured JSON after its arguments pass validation, useful for prototyping a contract before an implementation exists.

## Questions

### Where is my toolset saved?

The Save button writes one toolset to IndexedDB in the current browser. It is not uploaded to a workspace, and clearing site data removes it.

### Where does JavaScript execute?

Script tools execute locally in Postman Sandbox with cookies, vault, datasets, visualizer, and execution APIs disabled. Static-response tools return their configured JSON locally.

### What is sent to AI?

When you send a Playground prompt, selected tool names, descriptions, schemas, model-generated arguments, tool results, the system prompt, and conversation text participate in the request. The JavaScript source itself is not sent as a tool definition.

### Can I design MCP tools with this tool?

Yes. You can define MCP tool names, model-facing descriptions, JSON Schema inputs, and JavaScript or static-response implementations, then test how an AI calls them. This browser tool does not start an MCP server; a Postman toolset can add a local MCP server when you are ready to expose the tools to an MCP client.

### Does it start a local MCP server?

No. This signed-out browser tool does not expose a server, URL, port, client configuration, or server logs. In Postman, you can collaborate on and share the toolset, then run it as a local MCP server.


## Privacy

Toolset definitions and JavaScript are saved only in IndexedDB on this browser. JavaScript and static responses execute locally. AI Playground requests send conversation text, selected tool schemas, generated arguments, and tool results through the anonymous Postman AI gateway; clearing the chat removes conversation state from the page.

## Links

- [Use the tool](https://www.postman.com/tools/toolset-design) — the interactive version of this page.
- [Postman Tools](https://www.postman.com/tools) — every free tool in this set.
- [Toolsets documentation](https://learning.postman.com/docs/postman-ai-developer-tools/toolsets/overview/) — Postman documentation.