Creating a Tool

Tools in superglue are reusable workflows that you can execute on-demand, schedule, or trigger via webhooks. Each tool consists of sequential steps with built-in data transformations, loops, and error handling. Learn more about how tools fit into the bigger picture in our core concepts page.

Via Agent
Via UI
Via SDK

The fastest way to create a tool is by talking to our agent. Describe what you want to accomplish and the agent will build the tool for you.

Simple example:

"Create a tool that fetches all active customers from Stripe"

superglue will:

Identify which integration to use (or create one if needed)
Find the relevant API endpoint from the documentation
Configure the API call with appropriate parameters
Test it and make it available for execution

Complex example:

"Create a tool that loops through all contacts in HubSpot,
filters those with email domains matching @company.com,
and updates their lifecycle stage to 'customer'"

The agent can build multi-step tools with:

Sequential API calls
Data transformations between steps
Loops for batch processing
Error handling and retries

Try it now

Start building tools with our agent

For programmatic tool creation, use the SDK:

npm install @superglue/client

Build a tool with natural language:

SDK
GraphQL

import { SuperglueClient } from "@superglue/client";

const superglue = new SuperglueClient({
  apiKey: "your_api_key_here"
});

// Let superglue build the workflow from your instruction
const tool = await superglue.buildWorkflow({
  integrationIds: ["stripe"],
  instruction: "Fetch all active customers from Stripe"
});

console.log(tool); // Tool with auto-generated steps

mutation BuildWorkflow {
  buildWorkflow(
    integrationIds: ["stripe"]
    instruction: "Fetch all active customers from Stripe"
  ) {
    id
    instruction
    steps {
      id
      integrationId
      apiConfig {
        method
        urlPath
      }
    }
    finalTransform
  }
}

Build and execute a tool with natural language:

SDK
GraphQL

// Build the workflow from natural language
const tool = await superglue.buildWorkflow({
  integrationIds: ["hubspot"],
  instruction: `Get all contacts from HubSpot where the email domain is
                @company.com and update their lifecycle stage to 'customer'`
});

// Test it with sample data
const result = await superglue.executeWorkflow({
  workflow: tool,
  payload: {}
});

console.log(result); // { success: true, data: { updated: 5, contacts: [...] } }

// Save it for reuse
await superglue.upsertWorkflow({
  ...result.config,
  id: "update-company-contacts"
});

# Step 1: Build the workflow
mutation BuildWorkflow {
  buildWorkflow(
    integrationIds: ["hubspot"]
    instruction: """
    Get all contacts from HubSpot where the email domain is
    @company.com and update their lifecycle stage to 'customer'
    """
  ) {
    id
    steps {
      id
      integrationId
      apiConfig {
        method
        urlPath
        body
      }
    }
    finalTransform
  }
}

# Step 2: Execute the workflow
mutation ExecuteWorkflow($workflow: WorkflowInput!) {
  executeWorkflow(
    workflow: $workflow
    payload: {}
  ) {
    success
    data
    config {
      id
      steps {
        id
      }
    }
  }
}

# Step 3: Save for reuse
mutation SaveWorkflow($workflow: WorkflowInput!) {
  upsertWorkflow(workflow: $workflow) {
    id
    instruction
  }
}

See the SDK documentation for more details.

Tool anatomy

Every tool consists of:

Steps

Sequential API calls that fetch or modify data. Each step can reference data from previous steps.

Execution Mode

Steps run with self-healing enabled or disabled. Self-healing can auto-repair failures during execution.

Transformations

JavaScript functions that shape the step inputs and the final output. Ensures tool results adhere to response schemas.

Variables

Access data from previous steps, credentials, and input payload using <<variable>> syntax.

Step configuration

Each step in a tool represents a single API call with the following configuration:

Basic structure

{
  id: "stepId",                    // Unique identifier for this step
  integrationId: "stripe",         // Which integration to use
  loopSelector: "(sourceData) => { return [\"one\", \"two\"]}",  // The data selector for the current step
  apiConfig: {
    method: "POST",                 // HTTP method
    urlPath: "/v1/customers",      // API endpoint path
    queryParams: {},               // URL query parameters
    headers: { "Authorization": "Bearer <<stripe_apiKey>>" }, // Custom headers
    body: ""                       // Request body
  }
}

Variable syntax

Use <<variable>> to access dynamic values: Access credentials:

{
  headers: {
    "Authorization": "Bearer <<integrationId_access_token>>"
  }
}

Access previous step data:

{
  urlPath: "/customers/<<getCustomer.data.id>>",
  body: {
    email: "<<getCustomer.data.email>>"
  }
}

Access input payload:

{
  queryParams: {
    user_id: "<<userId>>",
    date: "<<startDate>>"
  }
}

Execute JavaScript expressions:

{
  body: {
    ids: "<<(sourceData) => JSON.stringify(sourceData.users.map(u => u.id))>>",
    timestamp: "<<(sourceData) => new Date().toISOString()>>",
    count: "<<(sourceData) => sourceData.items.length>>",
    uppercaseName: "<<(sourceData) => sourceData.name.toUpperCase()>>"
  }
}

JavaScript expressions must use the arrow function syntax (sourceData) => .... Direct property access like <<userId>> works for simple variables, but not for nested properties or transformations.

Data transformations

Data selector

Extract an array to iterate over:

{
  loopSelector: `(sourceData) => {
    const items = sourceData.fetchItems.results || [];
    const excludeIds = sourceData.excludeIds || [];
    return items.filter(item => !excludeIds.includes(item.id));
  }`;
}

Final transform

Shape the final output of the entire tool:

{
  finalTransform: `(sourceData) => {
    const customers = sourceData.getCustomers.data || [];
    const updated = sourceData.updateCustomers || [];
    
    return {
      success: true,
      total: customers.length,
      updated: updated.length,
      results: updated.map(r => ({
        id: r.currentItem.id,
        status: r.data.status
      }))
    };
  }`;
}

Special integrations

PostgreSQL

Query databases using the postgres:// URL scheme:

{
  id: "queryDatabase",
  integrationId: "postgres_db",
  apiConfig: {
    urlHost: "postgres://<<postgres_db_username>>:<<postgres_db_password>>@<<postgres_db_hostname>>:5432",
    urlPath: "<<postgres_db_database>>",
    body: {
      query: "SELECT * FROM users WHERE status = $1 AND created_at > $2",
      params: ["active", "<<(sourceData) => sourceData.startDate>>"]
    }
  }
}

Always use parameterized queries with $1, $2, etc. placeholders to prevent SQL injection. Provide values in the params array, which can include static values or <<>> expressions.

FTP/SFTP

Access files on FTP servers:

{
  id: "listFiles",
  integrationId: "ftp-server",
  apiConfig: {
    urlHost: "sftp://<<integrationId_username>>:<<integrationId_password>>@<<integrationId_hostname>>:22",
    urlPath: "/data",
    body: {
      operation: "list",
      path: "/reports"
    }
  }
}

Supported operations:

list - List directory contents
get - Download file (auto-parses CSV/JSON/XML)
put - Upload file
delete - Delete file
rename - Rename/move file
mkdir - Create directory
rmdir - Remove directory
exists - Check if file exists
stat - Get file metadata

Error handling

Automatic retries

Failed steps are automatically retried with exponential backoff

Self-healing

When enabled, superglue attempts to fix configuration errors automatically

Validation

Response data is validated against expected schemas

Graceful degradation

Handle missing data with optional chaining and defaults in transformations

Keep steps focused - Each step should make a single API call. Use transformations to prepare data, not additional steps.Use descriptive IDs - Step IDs are used to reference data in later steps. Use clear names like getCustomers, updateOrder, sendEmail.Avoid unnecessary loops - Don’t loop over thousands of items if the API supports batch operations. Check the documentation first.Test with real data - Test each step incrementally with production-like data before deploying.

Input and output schemas

Define schemas to make your tools more robust: Input schema - Validates payload before execution:

{
  "type": "object",
  "properties": {
    "userId": { "type": "string" },
    "startDate": { "type": "string", "format": "date" }
  },
  "required": ["userId"]
}

Response schema - Validates final output:

{
  "type": "object",
  "properties": {
    "success": { "type": "boolean" },
    "data": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "id": { "type": "string" },
          "email": { "type": "string" }
        }
      }
    }
  }
}

Getting Started

Guides

SDK Reference

API Reference

Try it now

Tool anatomy

Steps

Execution Mode

Transformations

Variables

Step configuration

Basic structure

Variable syntax

Data transformations

Data selector

Final transform

Special integrations

PostgreSQL

FTP/SFTP

Error handling

Automatic retries

Self-healing

Validation

Graceful degradation

Input and output schemas

Next steps

Debug your tool

Deploy to production

Getting Started

Guides

SDK Reference

API Reference

Try it now

​Tool anatomy

Steps

Execution Mode

Transformations

Variables

​Step configuration

​Basic structure

​Variable syntax

​Data transformations

​Data selector

​Final transform

​Special integrations

​PostgreSQL

​FTP/SFTP

​Error handling

Automatic retries

Self-healing

Validation

Graceful degradation

​Input and output schemas

​Next steps

Debug your tool

Deploy to production

Tool anatomy

Step configuration

Basic structure

Variable syntax

Data transformations

Data selector

Final transform

Special integrations

PostgreSQL

FTP/SFTP

Error handling

Input and output schemas

Next steps