A2A Protocol

Connecting via A2A Protocol

Cerebrus Pulse publishes an A2A Agent Card and answers A2A message/send requests (the A2A protocol, protocol version 0.3), so an A2A agent can find it and ask for a skill in plain language.

Agent Card Discovery

curl https://api.cerebruspulse.xyz/.well-known/agent-card.json

The card lists every skill with its price, the payment networks, and the A2A endpoint (https://api.cerebruspulse.xyz/a2a). The same card is served at /.well-known/agent.json and /a2a/.well-known/agent-card.json, and a copy is published at https://cerebruspulse.xyz/.well-known/agent-card.json.

Available Skills

Each skill is served, and priced, by one HTTP endpoint:

Skill ID	Served by	Price (USDC)	Needs a coin	Description
`confluence-analysis`	`GET /pulse/{coin}`	$0.025	Yes	Multi-timeframe technical analysis with the confluence vote share
`market-screener`	`GET /screener`	$0.06	No	Screen 50+ perpetuals by signal strength
`open-interest-analysis`	`GET /oi/{coin}`	$0.015	Yes	OI deltas, percentile, trend and price-OI divergence
`spread-analysis`	`GET /spread/{coin}`	$0.015	Yes	Impact spread and slippage estimates
`correlation-matrix`	`GET /correlation`	$0.05	No	BTC-alt correlation matrix with regime
`intelligence-bundle`	`GET /bundle/{coin}`	$0.05	Yes	Pulse, 24h funding and sentiment in one call
`basis-analysis`	`GET /basis/{coin}`	$0.02	Yes	Perp premium to the Hyperliquid oracle, and oracle vs Chainlink
`cex-dex-divergence`	`GET /cex-dex/{token}`	$0.02	Yes	Coinbase vs DEX/Chainlink divergence for one token
`collateral-health`	`GET /depeg`	$0.01	No	USDC peg status from Chainlink
`liquidation-heatmap`	`GET /liquidations/{coin}`	$0.03	Yes	Modelled liquidation zones with cascade risk
`market-stress`	`GET /arb`	$0.02	No	Cross-chain arbitrage stress index

Sending a Task

Send a JSON-RPC 2.0 message/send request to the A2A endpoint. This is free: it returns where and how to pay, not data.

import json
import urllib.request

body = json.dumps({
    "jsonrpc": "2.0",
    "method": "message/send",
    "id": "my-task-1",
    "params": {
        "message": {
            "role": "user",
            "parts": [{"type": "text", "text": "Analyze BTC confluence"}]
        }
    }
}).encode()

req = urllib.request.Request(
    "https://api.cerebruspulse.xyz/a2a",
    data=body,
    headers={"Content-Type": "application/json"},
    method="POST",
)

with urllib.request.urlopen(req, timeout=15) as resp:
    result = json.loads(resp.read())["result"]

quote = result["metadata"]["x402"]
print(result["status"]["state"])                   # input-required
print(quote["skill_id"], quote["amount"], quote["currency"])
print(quote["http"]["method"], quote["http"]["url"])  # pay here, with an x402 v2 client

The Reply to a Paid Skill

The task comes back in state input-required. Its message says, in words, what the skill costs and which URL to call; result.metadata.x402 says the same in fields:

Field	Description
`status`	`payment-required`
`skill_id`	The skill the request was matched to
`amount`	The price in USDC, e.g. `"0.025"` (the price of the endpoint that serves the skill)
`currency`	`USDC`
`networks`	The payment networks, each with `network`, `recipient` and `label` (Base, Solana)
`network`, `recipient`	The Base network and recipient, kept for older clients
`facilitator`	The x402 facilitator the gateway uses
`inBandPayment`	`false`: A2A messages cannot carry payment
`http.method`, `http.url`	The request to make: for example `GET https://api.cerebruspulse.xyz/pulse/BTC`
`http.route`	The route that serves the skill, e.g. `GET /pulse/:coin`
`http.paymentHeader`	`PAYMENT-SIGNATURE`: the header an x402 v2 client pays with
`paymentRequired`	The x402 v2 terms that endpoint’s 402 carries (absent if the gateway could not build them; the endpoint’s own 402 still has them)

Other outcomes:

No coin found for a skill that needs one: state input-required, asking which coin.
Service off, payments down, or a stale data feed: state failed, with the reason. Nothing is quoted.

Intent Extraction

Cerebrus Pulse matches the message to a skill by keywords, or you name the skill in metadata.skill_id.

Natural language

{
  "message": {
    "role": "user",
    "parts": [{"type": "text", "text": "Show me the top 10 strongest setups"}]
  }
}

This maps to market-screener. Without a match, the default is confluence-analysis.

Explicit skill ID

{
  "message": {
    "role": "user",
    "parts": [{"type": "text", "text": "SOL data"}],
    "metadata": {"skill_id": "spread-analysis"}
  }
}

Coin Extraction

For skills that need a coin, the symbol is taken from the message text: an upper-case symbol from /coins, or a name such as “ethereum” → ETH or “bitcoin” → BTC.

Task Retrieval

tasks/get returns a task the server stored. Replies to message/send are complete as sent and are not stored, so there is nothing to poll for; tasks/get on their ids reports “Task not found or expired”.

Rate Limits

The A2A endpoint shares the REST limits: 12 requests per minute per IP. See Rate Limits & Errors.