Skip to main content
The AI Builder is an AI assistant that builds complete connector configurations. It runs in your IDE (Claude Code, Cursor, VS Code) and uses 20+ specialized MCP tools to research APIs, generate YAML, validate syntax, and test actions.
The AI Builder is in early access. Contact [email protected] for credentials.

​
What It Does

Research APIs

Discovers provider documentation and endpoints automatically

Generate YAML

Creates complete connector configurations following the Falcon spec

Validate & Test

Validates syntax and tests actions against real APIs

MCP Testing

Conversational testing with automatic tool description improvements

​
Setup

​
Step 1: Install Dependencies

npm install -g @stackone/cli

​
Step 2: Fork the Template Repository

git clone https://github.com/YOUR-ORG/connectors-template.git
cd connectors-template
npm install

​
Step 3: Configure MCP Tools

npx @stackone/cli agent setup --local
This configures your IDE with the MCP server connection. The agent credentials are stored securely as environment variables.

​
Step 4: Start Building

Open your IDE and prompt: 
Build a Falcon connector for HiBob HRIS with employee management actions.

​
How It Works

The agent follows a strict 8-step workflow:
AI Builder workflow: Research → Version Validation → Config Building → YAML Validation → MCP Testing → Description Refinement → Security → Meta Feedback
  1. Research - Discovers API documentation and endpoints
  2. Version Validation - Resolves API version conflicts
  3. Config Building - Generates YAML with all actions
  4. YAML Validation - Runs stackone validate
  5. MCP Testing - Conversational testing via MCP skill
  6. Description Refinement - Improves tool descriptions based on test results
  7. Security - Scrambles credentials before storage
  8. Meta Feedback - Reports results for improvement

​
MCP Tools

The agent has access to 20+ specialized tools:

​
Research Tools

ToolDescription
discover_actionsAutonomous API discovery (5-15 min)
get_provider_actionsCheck S3 for indexed actions
vector_searchSemantic search StackOne knowledge base
web_searchWeb search for API documentation
fetchRetrieve content from URLs

​
Analysis Tools

ToolDescription
analyze_versioningDetect and resolve API version conflicts
get_external_integrationsFind how others integrate with the provider
scan_external_repoSearch open-source connector implementations
extract_oas_actionsParse OpenAPI specifications

​
Validation Tools

ToolDescription
test_actionsBatch test connector actions
scramble_credentialsSecure credentials before storage

​
StackOne Context

ToolDescription
get_stackone_categoriesAvailable categories (hris, ats, crm)
get_stackone_actionsUnified operations for a category
get_provider_coverageCurrent StackOne coverage for provider
get_docsStackOne documentation index

​
Agent Instructions

The agent’s behavior is defined in CLAUDE.md at your repository root. Key sections: Workflow: Strict 11-step process from research to completion Naming Conventions:
  • Provider folder: src/configs/provider-name/ (lowercase)
  • Config file: provider.connector.s1.yaml
Quality Standards:
  • Clear descriptions (1-2 sentences) for all operations
  • Full CRUD operations where applicable
  • Error handling and rate limiting
  • Conversational MCP testing for validation

CLAUDE.md - Agent Instructions

View the complete agent instructions and MCP tool documentation on GitHub.

​
Example Session

You:
Create a connector for HiBob HRIS with employee management actions
Agent:
I’ll start by researching the HiBob API. Launching action discovery… [Uses discover_actions("hibob")] [Uses get_stackone_actions("hris")] [Uses vector_search("authentication", "hibob")] Based on my research, HiBob uses Basic Auth with a service user ID and token. I’ve identified 15 relevant endpoints. Creating the connector configuration…
The agent creates:
  • src/configs/hibob/hibob.connector.s1.yaml (main file)
  • src/configs/hibob/hibob.employees.s1.partial.yaml
  • src/configs/hibob/hibob.time_off.s1.partial.yaml
Then validates, tests all actions, and reports completion.

​
MCP Testing Skill

The agent includes a built-in MCP testing skill for conversational validation:

MCP Testing Skill

View the test-mcp-connector skill that enables conversational testing with automatic tool description improvements.
 Key features:
  • Conversational testing - Test actions through natural language prompts
  • Tool description refinement - Automatically improves descriptions based on test results
  • Real API validation - Tests against actual provider endpoints

​
Troubleshooting

  • Run stackone agent setup --local again
  • Verify credentials are correct
  • Restart your IDE after setup
  • Check that .cursor/mcp.json or equivalent was created
  • Verify CLAUDE.md is at repository root
  • Check MCP configuration in IDE settings
  • Ensure the file follows the correct format
  • Run stackone validate <file> manually
  • Check src/configs/README.md for YAML structure
  • Ensure all required fields are present
  • Use --debug flag to see full request/response
  • Verify credentials are correct and not expired
  • Check action IDs match exactly (case-sensitive)

​
Next Steps

Connector Structure

Complete YAML configuration reference

CLI Reference

All CLI commands