AI-Assisted Development
WebroBot ships a first-class AI integration layer: an MCP (Model Context Protocol) server and a set of Claude Code skills designed for building pipelines, developing plugins, and administering the platform without leaving your editor.
Claude Code is our recommended environment for vibe coding. For technically demanding tasks — in particular the development of custom ETL and REST API plugins for technical partners — Claude Code with the /webrobot-plugin-dev skill is the most effective development experience. The skill carries full knowledge of the plugin SDK trait hierarchy, multi-tenancy patterns, Flyway conventions, and common pitfalls, making it possible to go from description to a working plugin in a single session.
Claude Code MCP Plugin
The webrobot-claude-plugin installs as a Claude Code plugin and exposes a set of domain skills that are aware of WebroBot's architecture, stage catalog, API schema, and multi-tenancy patterns.
Installing the Plugin
Follow the plugin installation instructions in your WebroBot admin panel under Settings → AI Integration. The plugin registers the MCP server endpoint and injects the skill definitions into your Claude Code environment.
Available Skills
Invoke a skill by typing its name in Claude Code. Each skill loads a domain-specific instruction set and has access to the MCP tools listed below.
/webrobot-admin
General administration: projects, jobs, executions, agents, datasets, cloud credentials, and LLM providers.
Use it to:
- Create and manage ETL projects and jobs
- Trigger and monitor pipeline executions
- Register cloud credentials (AWS, GCP, Azure)
- Configure LLM provider API keys (OpenAI, Groq, Anthropic, etc.)
- Browse datasets and execution logs
Example prompts:
/webrobot-admin list my active jobs for org 123
/webrobot-admin execute job "price-discovery" and follow the logs
/webrobot-admin show me the last 10 executions for project "catalog-etl"/webrobot-pipeline
Build and inspect pipeline YAML. The skill knows all native stages, connector stages, iextract, and Python Extensions.
Use it to:
- Generate complete pipeline YAML from a description
- Debug a failing pipeline by inspecting its stage structure
- Suggest the right stage for a given data source
- Review and improve an existing pipeline
Example prompts:
/webrobot-pipeline build a pipeline that loads products from PostgreSQL, visits each product_url, extracts the price and title, and saves to CSV
/webrobot-pipeline what args does load_db accept?
/webrobot-pipeline review this pipeline YAML and suggest improvements/webrobot-plugin-dev
Build ETL Scala plugins and REST API Jersey plugins. The skill knows the full SDK trait hierarchy, the ORM layer, Flyway migration patterns, and multi-tenancy requirements.
Use it to:
- Scaffold a new plugin project from scratch
- Implement a custom
WTransformStage,WSourceStage, orWAggregateStage - Write Flyway migration SQL with the correct naming convention
- Review a plugin implementation for multi-tenancy correctness
- Debug JDBC and ResultSet lifecycle issues
Example prompts:
/webrobot-plugin-dev create a WTransformStage that calls an external enrichment API and adds the result as new fields
/webrobot-plugin-dev write the Flyway migration to add a confidence_score column to my plugin's results table
/webrobot-plugin-dev review this sink stage for multi-tenancy issues/webrobot-python-extension
Create and register Python Extension functions. The skill knows all three registration modes and common field-type gotchas.
Use it to:
- Generate a Python function for a described transformation
- Register the function via the API
- Build a complete pipeline YAML that uses the registered function
- Debug why a function is dropping rows
Example prompts:
/webrobot-python-extension write a function that extracts a price from a raw string like "€12,99" and register it for org 123
/webrobot-python-extension create a URL normalizer that strips tracking parameters
/webrobot-python-extension why is my function dropping rows with missing prices?Cursor IDE Support
The MCP server exposes WebroBot admin tools directly in Cursor's AI assistant. Configure the server endpoint in your project's .cursorrules file or in Cursor's MCP settings panel.
Once configured, you can use natural language in the Cursor chat to create jobs, execute pipelines, and inspect results — all without switching to a browser or terminal.
.cursorrules example:
WebroBot MCP server: https://webrobot.example.com/mcp
Always scope DB queries by org_id from ctx.config("webrobot.org.id") in ETL stages.
Always use OrganizationContextHelper.getOrganizationId(req) in REST endpoints.
Never hardcode org IDs.MCP Tools Reference
These tools are available to all skills and directly in Cursor when the MCP server is configured.
| Tool | Description |
|---|---|
auth_me | Verify current auth token and return user/org context |
list_projects | List ETL projects for the current organization |
get_project | Get a single project by ID |
create_project | Create a new ETL project |
delete_project | Delete a project |
list_jobs | List jobs within a project |
get_job | Get a single job with its pipeline YAML |
create_job | Create a new job with a pipeline YAML definition |
delete_job | Delete a job |
execute_job | Trigger a job execution |
stop_job | Stop a running execution |
list_executions | List recent executions for a job or project |
get_execution_logs | Stream or fetch logs for a specific execution |
list_categories | List project categories |
list_agents | List AI agents configured in the platform |
get_agent | Get a single agent definition |
create_agent | Create a new AI agent |
get_agent_code | Get the code/prompt of an agent |
list_datasets | List output datasets for a project |
delete_dataset | Delete a dataset |
list_cloud_credentials | List registered cloud credentials (AWS, GCP, etc.) |
list_llm_providers | List configured LLM provider API keys |
llm_infer | Call an LLM provider directly (for agent logic and testing) |
Typical AI-Assisted Workflow
The following workflow describes how the skills compose in practice.
Step 1 — Scaffold a pipeline with /webrobot-pipeline
Describe what you want to build. The skill generates a complete pipeline YAML using the right stages for your data source and target.
/webrobot-pipeline I have a CSV of products with EAN and product_url columns.
I want to visit each URL, extract the price and stock status, and save the results
to a CSV. Build me the pipeline YAML.Step 2 — Execute and monitor with /webrobot-admin
Once the YAML looks right, use the admin skill to create a job and run it.
/webrobot-admin create a job in project "price-monitoring" with the YAML I just built,
name it "daily-price-scrape", then execute it and follow the logsStep 3 — Add a custom stage with /webrobot-plugin-dev
If the native stages don't cover a requirement (e.g. you need a UPSERT into a custom table with business logic), scaffold a plugin stage.
/webrobot-plugin-dev I need a WSinkStage that takes the extracted price and EAN,
looks up the product in the products table, and inserts a record into price_history
only if the price has changed from the last recorded value. Write the full stage class.Step 4 — Quick transforms with /webrobot-python-extension
For lightweight field transformations that don't justify a Scala plugin, use the Python Extension skill to generate and register a function.
/webrobot-python-extension the price field arrives as "€12.99 incl. VAT".
Write a Python function that extracts just the numeric value, registers it
for org 123, and gives me the pipeline YAML to use it.Tips for Effective AI-Assisted Pipeline Development
Be specific about field names. The AI skills produce better output when you name your fields explicitly: "the CSV has columns: ean, product_url, brand" vs "I have a CSV".
Iterate in small steps. Build one stage at a time. Verify each stage works before adding the next. Use get_execution_logs to inspect intermediate output.
Let the AI handle boilerplate. The ServiceLoader registration files, Flyway naming conventions, and multi-tenancy patterns are all things the plugin dev skill handles automatically — you describe the business logic.
Use Python Extensions for speed, plugins for reliability. Python Extensions are fast to create and iterate on. Once a transform is stable and needs to handle large volumes, consider porting it to a WTransformStage Scala plugin for better performance and type safety.