AI & Agentic Systems

OpenAPI → agent tools

Convert an OpenAPI spec into MCP or function-calling definitions — and see which operations probably shouldn't be on an agent's tool surface at all.

Converts OpenAPI operations into MCP or function-calling tool definitions — and then tells you which ones probably should not be exposed. Handing an agent every endpoint is the most common way to build a tool surface, and close to the worst.

OpenAPI (YAML or JSON)

openapi: 3.0.3
info: { title: Orders API, version: 1.0.0 }
paths:
  /orders:
    get:
      operationId: listOrders
      summary: List orders
      description: Returns orders for the authenticated account, newest first.
      parameters:
        - name: status
          in: query
          description: Filter by order status
          schema: { type: string, enum: [open, shipped, cancelled] }
      responses: { '200': { description: OK } }
    post:
      operationId: createOrder
      summary: Create an order
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              required: [sku, qty]
              properties:
                sku: { type: string, description: Product code }
                qty: { type: integer, description: Quantity ordered }
      responses: { '201': { description: Created } }
  /orders/{orderId}:
    delete:
      operationId: deleteOrder
      summary: Delete an order
      parameters:
        - name: orderId
          in: path
          required: true
          description: The order to delete
          schema: { type: string }
      responses: { '204': { description: Deleted } }
  /internal/admin/reindex:
    post:
      operationId: reindex
      summary: Rebuild the search index
      responses: { '202': { description: Accepted } }

MCP tools

Run the output through the tool linter before shipping it — generated descriptions are exactly the kind this catches. Then check what the whole set could do with the blast-radius calculator.

Request bodies are flattened into the argument object on purpose — models fill a flat argument list far better than a nested body. Local $refs are resolved; remote ones are marked rather than fetched, since resolving them would mean this page making network calls, which it does not. Runs entirely in your browser.

Other tools

These exist because the underlying problem is real. If the numbers you just put in look uncomfortable, that is usually worth a conversation.

Tell us what you're running