Unified Agent Protocol (UAP) – Core SDK Specification

Version: 1.0.0
Maintained by: WhoMeta Labs part of WhoMeta Inc.
License: Apache 2.0
Language: Python 3.10+
Repository: Private/internal (planned public release Q3 2025)

---

✨ Introduction

The Unified Agent Protocol (UAP) is a foundational interoperability layer designed to standardize the definition, registration, execution, and orchestration of AI agents and tools across diverse ecosystems.
UAP is not a runtime or a competing protocol like A2A or MCP – instead, it acts as a universal adapter, enabling seamless translation between heterogeneous agent formats, toolkits, and interface protocols.

The uap-core SDK is the reference Python implementation of this protocol, designed for SDK-level integration, automatic conversions, and full schema introspection.

---

🚀 Getting Started

Installation

Install UAP from PyPI using pip:

# Basic installation
pip install unified-agent-protocol

# With development tools
pip install unified-agent-protocol[dev]

# With all optional dependencies
pip install unified-agent-protocol[all]

Quick Start

1. Define a Simple Tool

from unifiedagentprotocol.models.tool import Tool, ToolParam

# Create a weather tool
weather_tool = Tool(
    name="get_weather",
    description="Get current weather for a city",
    parameters=[
        ToolParam(
            name="city",
            type="string",
            description="Target city name",
            required=True
        )
    ]
)

print(weather_tool.model_dump_json(indent=2))

2. Create an Agent

from unifiedagentprotocol.models.agent import Agent

# Create an agent with tools
weather_agent = Agent(
    name="WeatherBot",
    description="Provides weather information",
    version="1.0.0",
    tools=[weather_tool]
)

# Export to different formats
print("A2A Format:", weather_agent.to_a2a())
print("MCP Format:", weather_agent.to_mcp())
print("OpenAPI Spec:", weather_agent.to_openapi())

3. Parse Existing Tools

from unifiedagentprotocol.parser.openwebui import parse_openwebui_tool
from unifiedagentprotocol.parser.langchain import parse_langchain_tool

# Parse OpenWebUI tool definition
openwebui_json = {
    "name": "calculator",
    "description": "Basic calculator",
    # ... more fields
}
uap_tool = parse_openwebui_tool(openwebui_json)

# Parse LangChain tool
from langchain.tools import DuckDuckGoSearchRun
langchain_tool = DuckDuckGoSearchRun()
uap_tool = parse_langchain_tool(langchain_tool)

4. CLI Usage

# Convert OpenWebUI tools to MCP format
uap --input openwebui_tools.json --format mcp > mcp_tools.json

# Convert Swagger spec to UAP format
uap --input petstore.yaml --format uap > uap_tools.json

# Show help
uap --help

Requirements

• Python 3.10+
• Core dependencies: pydantic, typer, requests, jsonschema
• Optional: aiohttp (async support), orjson (performance), mkdocs (docs)

---

💡 Motivation

As the AI agent ecosystem evolves, developers face increasing friction when integrating tools across platforms like OpenWebUI, LangChain, Azure OpenAI Agents, OpenAPI-based agents, or proprietary agent chains.

Common challenges include:

• ❌ Fragmented agent and tool definition formats
• ❌ Missing bridges between proprietary agent runtimes
• ❌ Lack of universal abstraction for tool metadata, input types, and execution capabilities
• ❌ Friction when reusing agent definitions across platforms (e.g., MCP ↔ A2A ↔ OpenAPI)

UAP solves this by introducing a common schema + protocol that allows agents and tools to be described once – and deployed, registered, or bridged anywhere.

---

📦 Key Features (Milestone 1)

• 🧠 Unified JSON model: All agent, tool, trigger, and role definitions follow a strongly typed Pydantic schema.
• 🔌 Multi-source parsers:
  • parse_openwebui(json): Import tools from OpenWebUI format.
  • parse_langchain(tool): Extract tool metadata from LangChain definitions.
  • parse_openapi(spec): Map OpenAPI endpoints to UAP tools.
• 📤 Export bridges:
  • to_a2a(agent): Generate A2A-compatible payload.
  • to_mcp(agent): Convert to Model Context Protocol (MCP) schema.
  • to_openapi(tool): Derive standard OpenAPI spec from UAP tool.
• 🖥️ CLI (uap bind):
  • Run transformations via command-line: uap bind --input tools.json --format mcp
• 🛠️ Development-first SDK:
  • Works offline, no server required.
  • Fully typed Python models (intellisense, validation).
  • Optional integration with LangChain, FastAPI, and asyncio runtimes.

---

🧩 Core Concepts

Concept	Description
Tool	Describes an executable unit with input/output schemas and runtime hints.
Agent	A logical actor using one or more tools to fulfill a task or objective.
Trigger	Defines when and how agents/tools should activate (event, cron, intent).
Role	Describes access & behavioral context (e.g., "analyst", "investigator").
OutputSchema	Optional structure for results / downstream usage.
UIConfig	Describes how this entity is represented in GUIs (forms, widgets, prompts).

All objects are implemented as subclasses of pydantic.BaseModel and support:

• ✅ Full JSON validation
• ✅ .dict() / .json() / .from_json() compatibility
• ✅ Versioning fields
• ✅ Extension-safe typing (e.g., extra = "allow")

---

🔄 Ecosystem Bridges

Target Protocol	Bridge	Status	Description
A2A (Agent-to-Agent)	to_a2a()	✅	Convert UAP agent into valid A2A descriptor
MCP (Model Context Protocol)	to_mcp()	✅	Map UAP agent/tool into MCP-compliant schema
OpenAPI 3	to_openapi()	✅	Export UAP tool(s) as OpenAPI endpoints

These bridges allow inter-protocol operability – for example, developers can register a LangChain tool on OpenWebUI and then expose it in an A2A runtime via UAP translation.

---

📚 Example Use Case

# Convert OpenWebUI tools into MCP-ready format
uap bind --input tools_openwebui.json --format mcp > mcp_payload.json

---

🤝 Contributing

Contributions, issues and feature requests are very welcome!

1. Fork the repository
2. Create your feature branch (git checkout -b feat/awesome-feature)
3. Commit your changes (git commit -m 'feat: add awesome feature')
4. Push to the branch (git push origin feat/awesome-feature)
5. Open a pull request

For full guidelines, please read the CONTRIBUTE guide.

---

⚖️ License

This project is licensed under the Apache License 2.0 – see the LICENSE file for details.

---

📑 Changelog

All notable changes will be documented in CHANGELOG.md.