ProtoJSON

ProtoJSON is the canonical JSON encoding of Protocol Buffers messages, and A2A v1.0 adopts it as the normative way to write its data model as JSON in every JSON-based binding.

Rules that shape A2A payloads.

• Field names become lowerCamelCase: context_id in the proto is contextId in JSON. Parsers also accept the original snake_case name.
• Enums are written as the value’s name, so a task state is "TASK_STATE_COMPLETED" and a role is "ROLE_USER". Parsers also accept the integer.
• Oneof fields appear as whichever member is set, so the key identifies the type. That is why a v1.0 text part is an object with a text key and no kind field.
• Bytes become standard base64 strings, as in a part’s raw field.
• Timestamps become RFC 3339 strings. A2A narrows this to UTC with a Z suffix and recommends millisecond precision.
• Struct and Value hold arbitrary JSON, which is how metadata and data parts carry free-form content.
• Defaults of fields without presence are omitted, and serializers should not emit null.

A message in ProtoJSON, illustrative:

{
  "messageId": "0c4f1f7e-6a1b-4d2c-9d61-1f3b2a9e7c55",
  "role": "ROLE_USER",
  "parts": [{"text": "Where is order PO-5518?"}],
  "metadata": {"channel": "procurement"}
}

Why A2A chose it. ADR-001, accepted by the A2A Technical Steering Committee on 18 November 2025, picked ProtoJSON over writing custom transformation rules into the specification. The ADR lists the costs: a breaking change to enum casing, since v0.3 used lowercase strings such as completed, and no round-tripping of unknown fields.

A practical catch. ProtoJSON parsers reject unknown fields by default. Section 5.7 of the A2A specification says implementations should ignore unrecognized fields for forward compatibility. An A2A implementation built on a stock ProtoJSON parser therefore needs its ignore-unknown-fields option switched on.

Neighbouring terms. gRPC carries the same messages in binary protobuf. The JSON-RPC and HTTP+JSON bindings carry them as ProtoJSON.