🤖A2A (Agent-to-Agent)

The platform supports the A2A (Agent-to-Agent) protocol built on JSON-RPC 2.0. External agents can submit tasks, receive results in real time and interact with the system's specialised agents.

Architecture

Every request is handled by the Orchestrator — the top-level agent that analyses the task and delegates it to a specialised one:

Each agent works on the platform's real data through MCP tools.

Authentication

Every request needs an API token in the Authorization header. Create one in Settings → Connected Apps → API Tokens.

Authorization: Bearer sk_ws_<wsid>_<token>

Quick start

1. Fetch the Agent Card

curl https://stage.kompot.ai/.well-known/agent.json

Returns the agent description, its supported capabilities and skills.

2. Submit a task

curl -X POST https://stage.kompot.ai/api/ws/pasv/a2a \
  -H "Authorization: Bearer sk_ws_pasv_..." \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": "1",
    "method": "message/send",
    "params": {
      "message": {
        "role": "user",
        "parts": [
          {
            "kind": "text",
            "text": "Build an income statement for 2025"
          }
        ]
      }
    }
  }'

Response:

{
  "jsonrpc": "2.0",
  "id": "1",
  "result": {
    "id": "task-uuid",
    "status": "working",
    "messages": [...]
  }
}

3. Retrieve the result

curl -X POST https://stage.kompot.ai/api/ws/pasv/a2a \
  -H "Authorization: Bearer sk_ws_pasv_..." \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": "2",
    "method": "tasks/get",
    "params": {
      "id": "task-uuid"
    }
  }'

Response once the task is finished:

{
  "jsonrpc": "2.0",
  "id": "2",
  "result": {
    "id": "task-uuid",
    "status": "completed",
    "messages": [
      { "role": "user", "parts": [{ "kind": "text", "text": "..." }] },
      { "role": "agent", "parts": [{ "kind": "text", "text": "## Income Statement..." }] }
    ],
    "agentsUsed": ["orchestrator", "finance"]
  }
}

Methods

message/send

Submits a task to the agent. Execution is asynchronous — the method returns the task with status working, and the result is fetched with tasks/get.

message/stream

Submits a task and streams progress in real time over SSE. The response is text/event-stream carrying these events:

Example:

curl -N -X POST https://stage.kompot.ai/api/ws/pasv/a2a \
  -H "Authorization: Bearer sk_ws_pasv_..." \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": "1",
    "method": "message/stream",
    "params": {
      "message": {
        "role": "user",
        "parts": [{ "kind": "text", "text": "Show me every unpaid invoice" }]
      }
    }
  }'

tasks/get

Fetches the status and result of a task by its ID.

Task statuses:

Connecting through Claude Code

The platform is also available as an MCP server. Claude Code can use the ask_agent tool to talk to the agents directly over an existing MCP connection.

MCP configuration in .mcp.json:

{
  "mcpServers": {
    "kompot": {
      "type": "http",
      "url": "https://stage.kompot.ai/api/ws/pasv/mcp/stream",
      "headers": {
        "Authorization": "Bearer sk_ws_pasv_..."
      }
    }
  }
}

Example tasks