Usage Guide

This page covers common workflows for both the CLI and the Flask Web UI.
For installation instructions see Installation.

1  Command‑line interface

1.1 Basic syntax

agent-generator [OPTIONS] "plain‑English requirement"

1.2 Frequently used flags

Flag / Option	Description	Example
-f, --framework *	Which generator to use (crewai, langgraph, …).	--framework crewai
-p, --provider	LLM back‑end (watsonx default, or openai).	--provider openai
--model	Override default model for the provider.	--model gpt-4o
--temperature	Sampling randomness (0–2).	--temperature 0.3
--max-tokens	Response length cap.	--max-tokens 2048
--mcp / --no-mcp	Wrap Python output in an MCP FastAPI server.	--mcp
-o, --output PATH	Write result to file instead of stdout.	-o team.py
--dry-run	Build workflow + code skeleton but skip LLM call.	--dry-run
--show-cost	Print token counts & approximate USD cost.	--show-cost

1.3 Common recipes

Goal	Command
Orchestrate YAML from one‑liner	agent-generator "Email summariser" -f watsonx_orchestrate -o summariser.yaml
CrewAI Flow with MCP wrapper	agent-generator "Analyse tweets" -f crewai_flow --mcp -o tweets_flow.py
Cost estimate only	agent-generator "Scrape website" -f react --dry-run --show-cost
Use OpenAI instead of WatsonX	agent-generator "Write jokes" -f react -p openai --model gpt-4o

2  Flask Web UI

2.1 Run locally

FLASK_APP=agent_generator.web FLASK_ENV=development flask run
# visit http://localhost:5000

2.2 Workflow

1. Fill in prompt – describe your requirement.
2. Pick framework & provider – drop‑downs.
3. (Optional) toggle MCP wrapper.
4. Click Generate.
5. Download the code/YAML or copy‑paste from the preview.
6. Mermaid diagram appears under the code for quick validation.

3  Docker usage

docker build -t agent-generator .
docker run -e WATSONX_API_KEY=... -e WATSONX_PROJECT_ID=... \
           -p 8000:8000 agent-generator
# Web UI → http://localhost:8000

You can also exec into the container to run the CLI:

docker run --rm agent-generator agent-generator "Say hi" -f react --dry-run

4  Serving generated MCP skills

Every Python framework (crewai, crewai_flow, langgraph, react) can be generated with an MCP wrapper:

agent-generator "...data pipeline..." -f langgraph --mcp -o pipeline.py
python pipeline.py serve      # exposes POST /invoke on port 8080

Upload the packaged script or its Docker image to your MCP Gateway and then import it as a custom skill in WatsonX Orchestrate.

5  Troubleshooting

Symptom	Resolution
CLI raises 401 (WatsonX)	Verify WATSONX_API_KEY, WATSONX_PROJECT_ID, region URL.
ModuleNotFoundError: flask	pip install "agent-generator[web]"
Diagram doesn’t render in UI	Check browser console – Mermaid JS must load (make sure unpkg.com isn’t blocked).
High cost estimate	Lower --max-tokens or pick llama‑3‑8b instead.
Gateway import fails	Ensure you used --mcp and port 8080 is exposed.

Still stuck? Open an issue on the GitHub tracker.

Jump in: Installation ➜ · Usage ➜ · Frameworks ➜