mirror of
https://github.com/crewAIInc/crewAI.git
synced 2026-01-10 16:48:30 +00:00
feat: Add A2A (Agent-to-Agent) protocol support for remote interoperability
- Implement CrewAgentExecutor class that wraps CrewAI crews as A2A-compatible agents - Add server utilities for starting A2A servers with crews - Include comprehensive test coverage for all A2A functionality - Add optional dependency group 'a2a' in pyproject.toml - Expose A2A classes in main CrewAI module with graceful import handling - Add documentation and examples for A2A integration - Support bidirectional agent communication via A2A protocol - Enable crews to participate in remote agent networks Fixes #2970 Co-Authored-By: João <joao@crewai.com>
This commit is contained in:
Devin AI
184
docs/concepts/a2a-integration.md
Normal file
184
docs/concepts/a2a-integration.md
Normal file
@@ -0,0 +1,184 @@
|
||||
# A2A Protocol Integration
|
||||
|
||||
CrewAI supports the A2A (Agent-to-Agent) protocol, enabling your crews to participate in remote agent interoperability. This allows CrewAI crews to be exposed as remotely accessible agents that can communicate with other A2A-compatible systems.
|
||||
|
||||
## Overview
|
||||
|
||||
The A2A protocol is Google's standard for agent interoperability that enables bidirectional communication between agents. CrewAI's A2A integration provides:
|
||||
|
||||
- **Remote Interoperability**: Expose crews as A2A-compatible agents
|
||||
- **Bidirectional Communication**: Enable full-duplex agent interactions
|
||||
- **Protocol Compliance**: Full support for A2A specifications
|
||||
- **Transport Flexibility**: Support for multiple transport protocols
|
||||
|
||||
## Installation
|
||||
|
||||
A2A support is available as an optional dependency:
|
||||
|
||||
```bash
|
||||
pip install crewai[a2a]
|
||||
```
|
||||
|
||||
## Basic Usage
|
||||
|
||||
### Creating an A2A Server
|
||||
|
||||
```python
|
||||
from crewai import Agent, Crew, Task
|
||||
from crewai.a2a import CrewAgentExecutor, start_a2a_server
|
||||
|
||||
# Create your crew
|
||||
agent = Agent(
|
||||
role="Assistant",
|
||||
goal="Help users with their queries",
|
||||
backstory="A helpful AI assistant"
|
||||
)
|
||||
|
||||
task = Task(
|
||||
description="Help with: {query}",
|
||||
agent=agent
|
||||
)
|
||||
|
||||
crew = Crew(agents=[agent], tasks=[task])
|
||||
|
||||
# Create A2A executor
|
||||
executor = CrewAgentExecutor(crew)
|
||||
|
||||
# Start A2A server
|
||||
start_a2a_server(executor, host="0.0.0.0", port=10001)
|
||||
```
|
||||
|
||||
### Custom Configuration
|
||||
|
||||
```python
|
||||
from crewai.a2a import CrewAgentExecutor, create_a2a_app
|
||||
|
||||
# Create executor with custom content types
|
||||
executor = CrewAgentExecutor(
|
||||
crew=crew,
|
||||
supported_content_types=['text', 'application/json', 'image/png']
|
||||
)
|
||||
|
||||
# Create custom A2A app
|
||||
app = create_a2a_app(
|
||||
executor,
|
||||
agent_name="My Research Crew",
|
||||
agent_description="A specialized research and analysis crew",
|
||||
transport="starlette"
|
||||
)
|
||||
|
||||
# Run with custom ASGI server
|
||||
import uvicorn
|
||||
uvicorn.run(app, host="0.0.0.0", port=8080)
|
||||
```
|
||||
|
||||
## Key Features
|
||||
|
||||
### CrewAgentExecutor
|
||||
|
||||
The `CrewAgentExecutor` class wraps CrewAI crews to implement the A2A `AgentExecutor` interface:
|
||||
|
||||
- **Asynchronous Execution**: Crews run asynchronously within the A2A protocol
|
||||
- **Task Management**: Automatic handling of task lifecycle and cancellation
|
||||
- **Error Handling**: Robust error handling with A2A-compliant responses
|
||||
- **Output Conversion**: Automatic conversion of crew outputs to A2A artifacts
|
||||
|
||||
### Server Utilities
|
||||
|
||||
Convenience functions for starting A2A servers:
|
||||
|
||||
- `start_a2a_server()`: Quick server startup with default configuration
|
||||
- `create_a2a_app()`: Create custom A2A applications for advanced use cases
|
||||
|
||||
## Protocol Compliance
|
||||
|
||||
CrewAI's A2A integration provides full protocol compliance:
|
||||
|
||||
- **Agent Cards**: Automatic generation of agent capability descriptions
|
||||
- **Task Execution**: Asynchronous task processing with event queues
|
||||
- **Artifact Management**: Conversion of crew outputs to A2A artifacts
|
||||
- **Error Handling**: A2A-compliant error responses and status codes
|
||||
|
||||
## Use Cases
|
||||
|
||||
### Remote Agent Networks
|
||||
|
||||
Expose CrewAI crews as part of larger agent networks:
|
||||
|
||||
```python
|
||||
# Multi-agent system with specialized crews
|
||||
research_crew = create_research_crew()
|
||||
analysis_crew = create_analysis_crew()
|
||||
writing_crew = create_writing_crew()
|
||||
|
||||
# Expose each as A2A agents on different ports
|
||||
start_a2a_server(CrewAgentExecutor(research_crew), port=10001)
|
||||
start_a2a_server(CrewAgentExecutor(analysis_crew), port=10002)
|
||||
start_a2a_server(CrewAgentExecutor(writing_crew), port=10003)
|
||||
```
|
||||
|
||||
### Cross-Platform Integration
|
||||
|
||||
Enable CrewAI crews to work with other agent frameworks:
|
||||
|
||||
```python
|
||||
# CrewAI crew accessible to other A2A-compatible systems
|
||||
executor = CrewAgentExecutor(crew)
|
||||
start_a2a_server(executor, host="0.0.0.0", port=10001)
|
||||
|
||||
# Other systems can now invoke this crew remotely
|
||||
```
|
||||
|
||||
## Advanced Configuration
|
||||
|
||||
### Custom Agent Cards
|
||||
|
||||
```python
|
||||
from a2a.types import AgentCard, AgentCapabilities, AgentSkill
|
||||
|
||||
# Custom agent card for specialized capabilities
|
||||
agent_card = AgentCard(
|
||||
name="Specialized Research Crew",
|
||||
description="Advanced research and analysis capabilities",
|
||||
version="2.0.0",
|
||||
capabilities=AgentCapabilities(
|
||||
streaming=True,
|
||||
pushNotifications=False
|
||||
),
|
||||
skills=[
|
||||
AgentSkill(
|
||||
id="research",
|
||||
name="Research Analysis",
|
||||
description="Comprehensive research and analysis",
|
||||
tags=["research", "analysis", "data"]
|
||||
)
|
||||
]
|
||||
)
|
||||
```
|
||||
|
||||
### Error Handling
|
||||
|
||||
The A2A integration includes comprehensive error handling:
|
||||
|
||||
- **Validation Errors**: Input validation with clear error messages
|
||||
- **Execution Errors**: Crew execution errors converted to A2A artifacts
|
||||
- **Cancellation**: Proper task cancellation support
|
||||
- **Timeouts**: Configurable timeout handling
|
||||
|
||||
## Best Practices
|
||||
|
||||
1. **Resource Management**: Monitor crew resource usage in server environments
|
||||
2. **Error Handling**: Implement proper error handling in crew tasks
|
||||
3. **Security**: Use appropriate authentication and authorization
|
||||
4. **Monitoring**: Monitor A2A server performance and health
|
||||
5. **Scaling**: Consider load balancing for high-traffic scenarios
|
||||
|
||||
## Limitations
|
||||
|
||||
- **Optional Dependency**: A2A support requires additional dependencies
|
||||
- **Transport Support**: Currently supports Starlette transport only
|
||||
- **Synchronous Crews**: Crews execute synchronously within async A2A context
|
||||
|
||||
## Examples
|
||||
|
||||
See the `examples/a2a_integration_example.py` file for a complete working example of A2A integration with CrewAI.
|
||||
Reference in New Issue
Block a user