[FEATURE] Public Converters for Protocol Interoperability

[mkmeral]

Problem Statement

Strands Agents has internal converters for translating between Strands types and various external protocols (A2A, OpenAI, etc.), but these are currently private implementation details. Developers building custom integrations, middleware, or extending SDK functionality need access to these conversion utilities.

Current private converters:

• A2A: _converters.py - converts between Strands ContentBlocks/Messages and A2A Parts/Messages
• OpenAI: Internal conversions in model providers for OpenAI-compatible formats
• Bedrock: Internal conversions for Bedrock Converse API format

Without public converters, developers must:

• Duplicate conversion logic in their own code
• Reverse-engineer internal implementations
• Risk breaking changes when internals change

Proposed Solution

Create a public strands.converters module (or per-protocol submodules) exposing well-documented conversion functions:

from strands.converters import a2a, openai

# A2A conversions
a2a_message = a2a.to_message(strands_input)
strands_result = a2a.to_agent_result(a2a_response)
a2a_parts = a2a.to_parts(content_blocks)
content_blocks = a2a.from_parts(a2a_parts)

# OpenAI conversions
openai_messages = openai.to_messages(strands_messages)
strands_messages = openai.from_messages(openai_messages)
openai_tools = openai.to_tools(tool_specs)

Scope

Protocol	Direction	Priority
A2A	Strands ↔ A2A Messages/Parts	High
OpenAI	Strands ↔ OpenAI chat format	High
Bedrock	Strands ↔ Converse API format	Medium
Anthropic	Strands ↔ Anthropic API format	Medium

Design Considerations

1. Bidirectional: Support both to_* and from_* conversions where applicable
2. Lossless where possible: Preserve metadata and content fidelity
3. Clear error handling: Raise informative errors for unsupported content types
4. Versioned: Consider protocol version compatibility (e.g., OpenAI API versions)

Use Case

1. Custom A2A clients: Building A2A integrations beyond what A2AAgent provides
2. OpenAI-compatible servers: Exposing Strands agents via OpenAI-compatible APIs
3. Protocol bridges: Converting between different agent protocols
4. Testing/debugging: Inspecting converted payloads during development
5. Middleware: Building request/response interceptors that need to understand both formats

Alternatives Solutions

No response

Additional Context

No response