Build AI agent applications with the Claude Agent SDK.
Production-ready backend + frontend in minutes.
Preview
5 different agents, one framework. Click to explore.
Built-in Features
9 features. All optional. Enable what you need, skip what you don't.
Persist facts, preferences, and feedback across sessions
Learn document styles, create reusable workflows with parameters
Run prompts on cron schedules — daily briefs, weekly reports
Agent creates and executes Python functions at runtime
Generate presentations, reports, and documents from structured data
Connect external MCPs — Gmail, Slack, Calendar, Jira
Record user ratings and route to Langfuse analytics
Log every tool call with timing to PostgreSQL
Persist chat conversations across page reloads
Define a tool, wrap it in a Feature, configure, and go.
# 1. Define a tool
@tool("lookup", "Look up a customer.", {"name": str})
async def lookup(args):
return {"content": [{"type": "text", "text": f"Found: {args['name']}"}]}
# 2. Wrap it in a Feature
crm_feature = Feature(
name="crm",
mcp_server_factory=lambda **kw: create_sdk_mcp_server("crm", tools=[lookup]),
sidebar_tab={"key": "crm", "label": "CRM", "icon": "users"},
)
# 3. Configure and create
config = FrameworkConfig(
app_name="Support Bot",
system_prompt="You are a support assistant.",
db_dsn=None,
)
app = create_app(config, features=[crm_feature])Architecture
git clone github.com/hfarazul/claude-forge
pip install -e ".[dev]"
cd frontend && npm installcp .env.example .env
# Set ANTHROPIC_API_KEYuvicorn backend.main:app --port 8080
cd frontend && npm run dev -- -p 3001| Layer | Technology |
|---|---|
| Backend | FastAPI + Claude Agent SDK (Python 3.12) |
| Frontend | Next.js 16 + React 19 + Vercel AI SDK |
| Database | PostgreSQL via asyncpg (optional) |
| Streaming | SSE — Vercel AI SDK UI Message Stream Protocol v1 |
| Testing | pytest (194 backend) + vitest (35 frontend) |
Open source. MIT licensed. Build something extraordinary.