bidur

Bi dur, some info is missing! Let me generate UI to gather the missing pieces.

Dynamic form generation for AI agents via Model Context Protocol (MCP).

Live Demo

Try bidur Demo →

What It Does

Agent: "I need api_key, project_name from the user"
       ↓
MCP Server creates a form → User fills it in browser → Data returns to agent

Quick Start

Option 1: Remote SSE (Recommended for Testing)

from agents import Agent, Runner
from agents.mcp import MCPServerSse

async def main():
    async with MCPServerSse(
        params={
            "url": "https://gen-ui-mcp-server.onrender.com/sse",
            "headers": {"X-OpenAI-API-Key": "sk-..."},
        },
        client_session_timeout_seconds=300,
    ) as mcp_server:
        
        agent = Agent(
            name="My Agent",
            instructions="Use collect_user_input to gather info from user.",
            mcp_servers=[mcp_server],
        )
        
        result = await Runner.run(agent, "Collect user's name and email")

Option 2: Docker (Local Development)

cd docker
docker-compose up -d

# MCP Server: http://localhost:8080
# Form Server: http://localhost:9110

Then test with MCP Inspector:

npx @modelcontextprotocol/inspector

Set URL to http://localhost:8080/sse in browser UI. Add a new Custom Header named X-OpenAI-API-Key. Enter your OpenAI API key in the Value field.

Option 3: Remote SSE with MCP Inspector

npx @modelcontextprotocol/inspector sse https://gen-ui-mcp-server.onrender.com/sse --header "X-OpenAI-API-Key: sk-..."

Then in the browser UI:

1. Set URL to: https://gen-ui-mcp-server.onrender.com/sse
2. Go to Tools tab
3. Run collect_user_input with test input
4. Check Server Notifications for form URL

Testing

python examples/mcp_sse_example.py

Installation

git clone https://github.com/ogulcanakca/bidur.git
cd bidur
pip install -r requirements.txt

export OPENAI_API_KEY=sk-...

Available Tool

collect_user_input

Parameter	Type	Description
missing_fields	list[str]	Field names to collect (e.g., ["username", "email"])
context	str	Optional context for better form generation
timeout_seconds	int	Max wait time (default: 300)

Returns:

{
  "form_url": "https://...",
  "message": "User successfully submitted the form",
  "data": {"username": "john", "email": "john@example.com"}
}

Architecture

┌─────────────┐     SSE      ┌─────────────┐     HTTP     ┌─────────────┐
│   Agent     │◄────────────►│ MCP Server  │◄────────────►│ Form Server │
│ (Your App)  │              │  Port 8080  │              │  Port 9110  │
└─────────────┘              └─────────────┘              └─────────────┘
                                    │                            │
                              Tool Handler                  HTML/JS Form
                              API Key Mgmt                  Schema Gen

Project Structure

gen-ui-mcp/
├── src/gen_ui/
│   ├── mcp_server/          # MCP Server implementation
│   │   ├── server.py        # SSE transport, tool routing
│   │   └── tools.py         # collect_user_input tool
│   ├── agents/              # Form schema generation
│   └── tools/               # Standalone tool (non-MCP)
├── ui/html/
│   ├── server.py            # Form Server HTTP API
│   └── index.html           # Form UI
├── docker/
│   ├── Dockerfile.mcp
│   ├── Dockerfile.form
│   └── docker-compose.yml
├── examples/
│   └── mcp_sse_example.py   # SSE connection example
└── run_mcp_server.py        # Entry point (stdio/sse)

Notes

Client Limitations

Research in Progress: VS Code, Cursor, and Claude Desktop clients do not currently display log notifications during tool execution. This means the form URL is not visible until after form submission. Investigation is ongoing for alternative approaches.

OpenAI Agents SDK Features

This project supports Guardrails and Tracing features from the OpenAI Agents SDK.

Built-in Guardrails:

Guardrail	Type	Description
safety_guardrail	Input	Validates field names, checks for injection patterns
llm_field_validation_guardrail	Input	LLM-based semantic validation for suspicious field names
schema_format_guardrail	Output	Validates JSON Schema structure and property definitions