Core Concepts - AdCP - Ad Context Protocol

Essential concepts for building AdCP clients, regardless of which protocol you’re using (MCP, A2A, or future protocols).

Task Status System

Every AdCP response includes a status field that tells you exactly what state the operation is in and what action you should take next.

Status Values

AdCP uses the same status values as the A2A protocol’s TaskState enum:

Status	Meaning	Your Action
`submitted`	Task queued for execution	Show “queued” indicator, wait for updates
`working`	Agent actively processing	Show progress, poll frequently for updates
`input-required`	Needs information from you	Read `message` field, prompt user, send follow-up
`completed`	Successfully finished	Process `data`, show success message
`canceled`	User/system canceled task	Show cancellation notice, clean up
`failed`	Error occurred	Show error from `message`, handle gracefully
`rejected`	Agent rejected the request	Show rejection reason, don’t retry
`auth-required`	Authentication needed	Prompt for auth, retry with credentials
`unknown`	Indeterminate state	Log for debugging, may need manual intervention

Response Structure

Every AdCP response has this structure:

{
  "status": "completed",           // Always present: what state we're in
  "message": "Found 5 products",  // Always present: human explanation
  "context_id": "ctx-123",         // Session continuity
  "data": {                        // Task-specific structured data
    "context": {                  // Application-level context echoed back
      "ui": "buyer_dashboard"
    },
    "products": [...]
  }
}

Client Decision Logic

Basic Status Handling

function handleAdcpResponse(response) {
  switch (response.status) {
    case 'completed':
      // Success - process the data
      showSuccess(response.message);
      return processData(response.data);
      
    case 'input-required':
      // Need more info - prompt user
      const userInput = await promptUser(response.message);
      return sendFollowUp(response.context_id, userInput);
      
    case 'working':
      // In progress - show progress and wait
      showProgress(response.message);
      return pollForUpdates(response.context_id);
      
    case 'failed':
      // Error - show message and handle gracefully
      showError(response.message);
      return handleError(response.data?.errors);
      
    case 'auth-required':
      // Authentication needed
      const credentials = await getAuth();
      return retryWithAuth(credentials);
      
    default:
      // Unexpected status
      console.warn('Unknown status:', response.status);
      showMessage(response.message);
  }
}

Advanced Status Patterns

1. Clarification Flow

When status is input-required, the message tells you what’s needed:

{
  "status": "input-required",
  "message": "I need more information about your campaign. What's your budget and target audience?",
  "context_id": "ctx-123",
  "data": {
    "products": [],  // Empty until clarification provided
    "suggestions": ["budget", "audience", "timing"]
  }
}

Client handling:

if (response.status === 'input-required') {
  // Extract what's needed from the message
  const missingInfo = extractRequirements(response.message);
  
  // Prompt user with specific questions
  const answers = await promptForInfo(missingInfo);
  
  // Send follow-up with same context_id
  return sendMessage(response.context_id, answers);
}

2. Approval Flow

Human approval is a special case of input-required:

{
  "status": "input-required", 
  "message": "Media buy exceeds auto-approval limit ($100K). Please approve to proceed with campaign creation.",
  "context_id": "ctx-123",
  "data": {
    "approval_required": true,
    "amount": 150000,
    "reason": "exceeds_limit"
  }
}

Client handling:

if (response.status === 'input-required' && response.data?.approval_required) {
  // Show approval UI
  const approved = await showApprovalDialog(response.message, response.data);
  
  // Send approval decision
  const decision = approved ? "Approved" : "Rejected";
  return sendMessage(response.context_id, decision);
}

3. Long-Running Operations

Async operations start with working and provide updates:

{
  "status": "working",
  "message": "Creating media buy. Validating inventory availability...",
  "context_id": "ctx-123", 
  "data": {
    "task_id": "task-456",
    "progress": 25,
    "step": "inventory_validation"
  }
}

Protocol-specific polling:

MCP: Poll with context_id for updates
A2A: Subscribe to SSE stream for real-time updates

Async Operations

Operation Types

AdCP operations fall into three categories:

Synchronous - Return immediately with completed or failed
- list_creative_formats, list_authorized_properties
- Fast operations that don’t require external systems
Interactive - May return input-required before proceeding
- get_products (when brief is vague or needs clarification)
- Operations that need user input to proceed
Asynchronous - Return working or submitted and require polling/streaming
- create_media_buy, activate_signal, sync_creatives, get_products
- Operations that integrate with external systems or require human approval

Timeout Handling

Set reasonable timeouts based on operation type:

const TIMEOUTS = {
  sync: 30_000,        // 30 seconds for immediate operations
  interactive: 300_000, // 5 minutes for human input
  working: 120_000,    // 2 minutes for working tasks
  submitted: 86_400_000 // 24 hours for submitted tasks
};

function setTimeoutForStatus(status) {
  switch (status) {
    case 'working': return TIMEOUTS.working;
    case 'submitted': return TIMEOUTS.submitted;
    case 'input-required': return TIMEOUTS.interactive;
    default: return TIMEOUTS.sync;
  }
}

Context Management

Session Continuity

The context_id maintains conversation state across requests:

class AdcpSession {
  constructor() {
    this.contextId = null;
  }
  
  async send(request) {
    // Include context from previous responses
    if (this.contextId) {
      request.context_id = this.contextId;
    }
    
    const response = await this.client.send(request);
    
    // Save context for next request
    this.contextId = response.context_id;
    
    return response;
  }
}

Context Expiration

Contexts typically expire after 1 hour of inactivity:

function isContextExpired(error) {
  return error.code === 'context_expired' || 
         error.message?.includes('context not found');
}

async function handleRequest(request) {
  try {
    return await session.send(request);
  } catch (error) {
    if (isContextExpired(error)) {
      // Start new conversation
      session.reset();
      return session.send(request);
    }
    throw error;
  }
}

Task Management & Webhooks

Task Tracking

All async operations return a task_id at the protocol level for tracking:

{
  "status": "submitted",
  "task_id": "task_456", 
  "message": "Media buy requires manual approval",
  "context_id": "ctx-123"
}

Push Notification Architecture

Both MCP and A2A use HTTP webhooks for async task updates. Instead of polling, you provide a webhook URL and the server POSTs status changes to you directly.

Aspect	MCP	A2A
Spec Status	AdCP specifies this	Native protocol feature
Configuration	`pushNotificationConfig`	`pushNotificationConfig`
Envelope	`mcp-webhook-payload.json`	`Task` or `TaskStatusUpdateEvent`
Data Location	`result` field	`status.message.parts[].data`
Data Schemas	Identical AdCP schemas	Identical AdCP schemas

MCP Webhooks

MCP doesn’t define push notifications. AdCP fills this gap by specifying the webhook configuration (pushNotificationConfig) and payload format (mcp-webhook-payload.json).

Note: If MCP adds native push notification support in future versions, AdCP will adopt that mechanism in a future major version to maintain alignment with the protocol’s evolution.

Envelope: mcp-webhook-payload.json
Data location: result field

const response = await session.call('create_media_buy',
  { /* task params */ },
  {
    pushNotificationConfig: {
      url: "https://buyer.com/webhooks/adcp",
      authentication: {
        schemes: ["HMAC-SHA256"],
        credentials: "shared_secret_32_chars"
      }
    }
  }
);

A2A Webhooks

A2A defines push notifications natively. Per the A2A spec, the server sends Task, TaskStatusUpdateEvent, or TaskArtifactUpdateEvent depending on what changed. Data location: status.message.parts[].data

await a2a.send({
  message: {
    parts: [{
      kind: "data",
      data: {
        skill: "create_media_buy",
        parameters: { /* task params */ }
      }
    }]
  },
  pushNotificationConfig: {
    url: "https://buyer.com/webhooks/a2a",
    authentication: {
      schemes: ["bearer"],
      credentials: "shared_secret_32_chars"
    }
  }
});

Unified Data Schemas

The data payload uses identical AdCP schemas regardless of envelope format:

MCP:  { result: { media_buy_id, packages, ... } }
A2A:  { status: { message: { parts: [{ data: { media_buy_id, packages, ... } }] } } }

Both validate against the same schemas. For completed/failed, use the full task response schema. For other statuses, use the status-specific schemas.

When Webhooks Are Called

Webhooks are triggered when all of the following are true:

Task type supports async execution (e.g., get_products, create_media_buy, sync_creatives)
pushNotificationConfig is provided in the request
Task requires async processing — initial response is working or submitted

If the initial response is already terminal (completed, failed, rejected), no webhook is sent — the client already has the final result. Status changes that trigger webhooks:

working → Progress update
input-required → Human input needed
completed → Final result available
failed → Error details
canceled → Cancellation confirmed

Push Notification Format by Protocol

MCP: HTTP Webhook POST

When an MCP async operation changes status, the publisher POSTs to your webhook URL. Envelope Schema: mcp-webhook-payload.json

POST /webhooks/adcp/create_media_buy/agent_123/op_456 HTTP/1.1
Host: buyer.example.com
Authorization: Bearer your-secret-token
Content-Type: application/json

{
  "task_id": "task_456",
  "task_type": "create_media_buy",
  "status": "completed",
  "timestamp": "2025-01-22T10:30:00Z",
  "message": "Media buy created successfully",
  "result": {
    "media_buy_id": "mb_12345",
    "buyer_ref": "nike_q1_campaign_2024",
    "creative_deadline": "2024-01-30T23:59:59Z",
    "packages": [
      { "package_id": "pkg_12345_001", "buyer_ref": "nike_ctv_sports_package" }
    ]
  }
}

A2A Webhook POST

A2A sends Task (for final states) or TaskStatusUpdateEvent (for progress updates):

{
  "id": "task_456",
  "contextId": "ctx_123",
  "status": {
    "state": "completed",
    "message": {
      "role": "agent",
      "parts": [
        { "text": "Media buy created successfully" },
        {
          "data": {
            "media_buy_id": "mb_12345",
            "buyer_ref": "nike_q1_campaign_2024",
            "creative_deadline": "2024-01-30T23:59:59Z",
            "packages": [
              { "package_id": "pkg_12345_001", "buyer_ref": "nike_ctv_sports_package" }
            ]
          }
        }
      ]
    },
    "timestamp": "2025-01-22T10:30:00Z"
  }
}

Status-Specific Data Schemas

The data payload (result in MCP, status.message.parts[].data in A2A) uses status-specific schemas:

Status	Data Schema	Contents
`completed`	`[task]-response.json`	Full task response (success branch)
`failed`	`[task]-response.json`	Full task response (error branch)
`working`	`[task]-async-response-working.json`	Progress info (`percentage`, `step`)
`input-required`	`[task]-async-response-input-required.json`	Requirements, approval data
`submitted`	`[task]-async-response-submitted.json`	Acknowledgment (usually minimal)

Supported Async Operations

Each async operation uses its main response schema for completed/failed statuses:

Task	Response Schema
`get_products`	`get-products-response.json`
`create_media_buy`	`create-media-buy-response.json`
`update_media_buy`	`update-media-buy-response.json`
`sync_creatives`	`sync-creatives-response.json`

MCP Webhook URL Patterns

For MCP HTTP webhooks, structure URLs to identify the operation:

https://buyer.com/webhooks/adcp/{task_name}/{agent_id}/{operation_id}

Your webhook handler can parse the URL path to route to the correct handler. Key principle: For async tasks with pushNotificationConfig, push notifications are sent for all status changes after the initial response. The data payload uses the same schema regardless of transport (MCP webhook or A2A native message).

Task State Reconciliation

Use tasks/list to recover from lost state:

// Find all pending operations
const pending = await session.call('tasks/list', {
  filters: {
    statuses: ["submitted", "working", "input-required"]
  }
});

// Reconcile with local state
const missingTasks = pending.tasks.filter(task => 
  !localState.hasTask(task.task_id)
);

// Resume tracking missing tasks
for (const task of missingTasks) {
  startPolling(task.task_id);
}

Status Progression

Tasks progress through predictable states:

submitted → working → completed
    ↓          ↓         ↑
input-required → → → → →
    ↓
  failed

submitted: Task queued for execution, provide webhook or poll
working: Agent actively processing, poll frequently
input-required: Need user input, continue conversation
completed: Success, process results
failed: Error, handle appropriately

For detailed timing expectations and polling patterns, see Task Management.

Webhook Reliability

Delivery Semantics

AdCP webhooks use at-least-once delivery semantics with the following characteristics:

Not guaranteed: Webhooks may fail due to network issues, server downtime, or configuration problems
May be duplicated: The same event might be delivered multiple times
May arrive out of order: Later events could arrive before earlier ones
Timeout behavior: Webhook delivery has limited retry attempts and timeouts

Security

Webhook Authentication (Required)

AdCP adopts A2A’s PushNotificationConfig structure for webhook configuration. This provides a standard, flexible authentication model that supports multiple security schemes. Configuration Structure (A2A-Compatible):

{
  "push_notification_config": {
    "url": "https://buyer.example.com/webhooks/adcp",
    "authentication": {
      "schemes": ["Bearer"],
      "credentials": "secret_token_min_32_chars"
    }
  }
}

Supported Authentication Schemes:

Bearer Token (Simple, Recommended for Development)

{
  "authentication": {
    "schemes": ["Bearer"],
    "credentials": "secret_token_32_chars"
  }
}

HMAC Signature (Enterprise, Recommended for Production)

{
  "authentication": {
    "schemes": ["HMAC-SHA256"],
    "credentials": "shared_secret_32_chars"
  }
}

Publisher Implementation (Bearer):

const config = pushNotificationConfig;
const scheme = config.authentication.schemes[0];

if (scheme === 'Bearer') {
  await axios.post(config.url, payload, {
    headers: {
      'Content-Type': 'application/json',
      'Authorization': `Bearer ${config.authentication.credentials}`
    }
  });
}

Publisher Implementation (HMAC-SHA256):

if (scheme === 'HMAC-SHA256') {
  const timestamp = new Date().toISOString();
  const signature = crypto
    .createHmac('sha256', config.authentication.credentials)
    .update(timestamp + JSON.stringify(payload))
    .digest('hex');

  await axios.post(config.url, payload, {
    headers: {
      'Content-Type': 'application/json',
      'X-ADCP-Signature': `sha256=${signature}`,
      'X-ADCP-Timestamp': timestamp
    }
  });
}

Buyer Implementation (Bearer):

app.post('/webhooks/adcp', async (req, res) => {
  const authHeader = req.headers.authorization;
  if (!authHeader || !authHeader.startsWith('Bearer ')) {
    return res.status(401).json({ error: 'Missing Authorization header' });
  }

  const token = authHeader.substring(7);
  if (token !== process.env.ADCP_WEBHOOK_TOKEN) {
    return res.status(401).json({ error: 'Invalid token' });
  }

  await processWebhook(req.body);
  res.status(200).json({ status: 'processed' });
});

Buyer Implementation (HMAC-SHA256):

app.post('/webhooks/adcp', async (req, res) => {
  const signature = req.headers['x-adcp-signature'];
  const timestamp = req.headers['x-adcp-timestamp'];

  if (!signature || !timestamp) {
    return res.status(401).json({ error: 'Missing signature headers' });
  }

  // Reject old webhooks (prevent replay attacks)
  const eventTime = new Date(timestamp);
  if (Date.now() - eventTime > 5 * 60 * 1000) {
    return res.status(401).json({ error: 'Webhook too old' });
  }

  // Verify signature
  const expectedSig = crypto
    .createHmac('sha256', process.env.ADCP_WEBHOOK_SECRET)
    .update(timestamp + JSON.stringify(req.body))
    .digest('hex');

  if (signature !== `sha256=${expectedSig}`) {
    return res.status(401).json({ error: 'Invalid signature' });
  }

  await processWebhook(req.body);
  res.status(200).json({ status: 'processed' });
});

Authentication Best Practices:

Bearer tokens: Simple, good for development and testing
HMAC signatures: Prevents replay attacks, recommended for production
Credentials exchanged out-of-band (during publisher onboarding)
Minimum 32 characters for all credentials
Store securely (environment variables, secret management)
Support credential rotation (accept old and new during transition)

Retry and Circuit Breaker Patterns

Publishers MUST implement retry logic with circuit breakers to handle temporary buyer endpoint failures without overwhelming systems or accumulating unbounded queues.

Retry Strategy

Publishers SHOULD use exponential backoff with jitter for webhook delivery retries:

class WebhookDelivery {
  constructor() {
    this.maxRetries = 3;
    this.baseDelay = 1000; // 1 second
    this.maxDelay = 60000; // 1 minute
  }

  async deliverWithRetry(url, payload, attempt = 0) {
    try {
      const response = await this.sendWebhook(url, payload);

      if (response.status >= 200 && response.status < 300) {
        return { success: true, attempts: attempt + 1 };
      }

      // Retry on 5xx errors and timeouts
      if (response.status >= 500 && attempt < this.maxRetries) {
        await this.delayWithJitter(attempt);
        return this.deliverWithRetry(url, payload, attempt + 1);
      }

      // Don't retry 4xx errors (client errors)
      return { success: false, error: 'Client error', attempts: attempt + 1 };

    } catch (error) {
      if (attempt < this.maxRetries) {
        await this.delayWithJitter(attempt);
        return this.deliverWithRetry(url, payload, attempt + 1);
      }
      return { success: false, error: error.message, attempts: attempt + 1 };
    }
  }

  async delayWithJitter(attempt) {
    const exponentialDelay = Math.min(
      this.baseDelay * Math.pow(2, attempt),
      this.maxDelay
    );
    // Add ±25% jitter to prevent thundering herd
    const jitter = exponentialDelay * (0.75 + Math.random() * 0.5);
    await new Promise(resolve => setTimeout(resolve, jitter));
  }

  async sendWebhook(url, payload) {
    return axios.post(url, payload, {
      timeout: 10000, // 10 second timeout
      headers: {
        'Content-Type': 'application/json',
        'X-ADCP-Signature': this.signPayload(payload),
        'X-ADCP-Timestamp': new Date().toISOString()
      }
    });
  }
}

Retry Schedule:

Attempt 1: Immediate
Attempt 2: After ~1 second (with jitter)
Attempt 3: After ~2 seconds (with jitter)
Attempt 4: After ~4 seconds (with jitter)
Give up after 4 total attempts

Circuit Breaker Pattern

Publishers MUST implement circuit breakers to prevent webhook queues from growing unbounded when buyer endpoints are down:

class CircuitBreaker {
  constructor(endpoint) {
    this.endpoint = endpoint;
    this.state = 'CLOSED'; // CLOSED, OPEN, HALF_OPEN
    this.failureCount = 0;
    this.failureThreshold = 5;
    this.successThreshold = 2;
    this.timeout = 60000; // 1 minute
    this.halfOpenTime = null;
    this.successCount = 0;
  }

  async execute(fn) {
    if (this.state === 'OPEN') {
      // Check if circuit should move to HALF_OPEN
      if (Date.now() - this.halfOpenTime > this.timeout) {
        this.state = 'HALF_OPEN';
        this.successCount = 0;
      } else {
        throw new Error('Circuit breaker is OPEN');
      }
    }

    try {
      const result = await fn();
      this.onSuccess();
      return result;
    } catch (error) {
      this.onFailure();
      throw error;
    }
  }

  onSuccess() {
    this.failureCount = 0;

    if (this.state === 'HALF_OPEN') {
      this.successCount++;
      if (this.successCount >= this.successThreshold) {
        this.state = 'CLOSED';
        console.log(`Circuit breaker CLOSED for ${this.endpoint}`);
      }
    }
  }

  onFailure() {
    this.failureCount++;

    if (this.failureCount >= this.failureThreshold) {
      this.state = 'OPEN';
      this.halfOpenTime = Date.now();
      console.error(`Circuit breaker OPEN for ${this.endpoint}`);

      // Alert monitoring system
      this.alertMonitoring();
    }
  }

  alertMonitoring() {
    // Notify operations team that endpoint is down
    console.error(`ALERT: Webhook endpoint ${this.endpoint} is unreachable`);
    // Send to monitoring system (e.g., PagerDuty, Datadog)
  }

  isOpen() {
    return this.state === 'OPEN';
  }
}

// Usage with webhook delivery
class WebhookManager {
  constructor() {
    this.circuitBreakers = new Map();
    this.maxQueueSize = 1000; // Per endpoint
    this.queues = new Map();
  }

  getCircuitBreaker(endpoint) {
    if (!this.circuitBreakers.has(endpoint)) {
      this.circuitBreakers.set(endpoint, new CircuitBreaker(endpoint));
    }
    return this.circuitBreakers.get(endpoint);
  }

  async sendWebhook(endpoint, payload) {
    const breaker = this.getCircuitBreaker(endpoint);

    // Check circuit breaker before queuing
    if (breaker.isOpen()) {
      console.warn(`Dropping webhook for ${endpoint} - circuit breaker OPEN`);
      return { success: false, reason: 'circuit_breaker_open' };
    }

    // Check queue size limit
    const queue = this.queues.get(endpoint) || [];
    if (queue.length >= this.maxQueueSize) {
      console.error(`Dropping webhook for ${endpoint} - queue full (${queue.length})`);
      return { success: false, reason: 'queue_full' };
    }

    // Attempt delivery through circuit breaker
    try {
      return await breaker.execute(async () => {
        const delivery = new WebhookDelivery();
        return await delivery.deliverWithRetry(endpoint, payload);
      });
    } catch (error) {
      return { success: false, reason: error.message };
    }
  }
}

Circuit Breaker States:

CLOSED: Normal operation, webhooks delivered
OPEN: Endpoint is down, webhooks are dropped (not queued)
HALF_OPEN: Testing if endpoint recovered, limited webhooks sent

Why Circuit Breakers Matter: At Yahoo scale with thousands of campaigns, a single buyer endpoint being down could queue millions of webhooks. Circuit breakers prevent this by failing fast and dropping webhooks when endpoints are unreachable.

Queue Management

Publishers SHOULD implement bounded queues with overflow policies:

class BoundedWebhookQueue {
  constructor(maxSize = 1000) {
    this.maxSize = maxSize;
    this.queue = [];
    this.droppedCount = 0;
  }

  enqueue(webhook) {
    if (this.queue.length >= this.maxSize) {
      // Overflow policy: drop oldest webhooks
      const dropped = this.queue.shift();
      this.droppedCount++;
      console.warn(`Dropped webhook ${dropped.id} due to queue overflow`);
    }
    this.queue.push(webhook);
  }

  dequeue() {
    return this.queue.shift();
  }

  size() {
    return this.queue.length;
  }

  getDroppedCount() {
    return this.droppedCount;
  }
}

Best Practices:

Set max queue size based on available memory and recovery time
Monitor queue depth and dropped webhook counts
Alert operations when queues are consistently full
Use dead letter queues for manual investigation of persistent failures
Implement queue per buyer endpoint (not global queue)

Implementation Requirements

Idempotent Webhook Handlers

Always implement idempotent webhook handlers that can safely process the same event multiple times:

app.post('/webhooks/adcp', async (req, res) => {
  const { task_id, current_status, timestamp, event_id } = req.body;
  
  // Idempotent check - avoid duplicate processing
  const existing = await db.getWebhookEvent(event_id);
  if (existing) {
    console.log(`Webhook ${event_id} already processed`);
    return res.status(200).json({ status: 'already_processed' });
  }
  
  // Record this webhook event
  await db.recordWebhookEvent(event_id, timestamp);
  
  // Process the status change
  await processTaskStatusChange(task_id, current_status, timestamp);
  
  // Always return 200 for successful processing
  res.status(200).json({ status: 'processed' });
});

Sequence Handling

Use timestamps to ensure proper event ordering:

async function processTaskStatusChange(taskId, newStatus, timestamp) {
  const currentTask = await db.getTask(taskId);
  
  // Ignore out-of-order events
  if (currentTask?.updated_at >= timestamp) {
    console.log(`Ignoring out-of-order webhook for task ${taskId}`);
    return;
  }
  
  // Update task with new status
  await db.updateTask(taskId, {
    status: newStatus,
    updated_at: timestamp
  });
  
  // Trigger any business logic
  await handleStatusChange(taskId, newStatus);
}

Polling as Backup

Use polling as a reliable backup mechanism:

class TaskTracker {
  constructor() {
    this.pendingTasks = new Map();
    this.pollInterval = 30000; // 30 seconds
  }
  
  async trackTask(taskId, webhookConfigured = false) {
    this.pendingTasks.set(taskId, {
      lastPolled: Date.now(),
      webhookConfigured,
      pollAttempts: 0
    });
    
    // Start polling backup even if webhook is configured
    this.schedulePolling(taskId);
  }
  
  async schedulePolling(taskId) {
    const task = this.pendingTasks.get(taskId);
    if (!task) return;
    
    // Increase polling interval if webhook is configured
    const interval = task.webhookConfigured ? 
      this.pollInterval * 4 : // 2 minutes with webhook
      this.pollInterval;      // 30 seconds without webhook
    
    setTimeout(async () => {
      if (this.pendingTasks.has(taskId)) {
        await this.pollTask(taskId);
        this.schedulePolling(taskId); // Continue polling
      }
    }, interval);
  }
  
  async pollTask(taskId) {
    try {
      const response = await adcp.call('tasks/get', {
        task_id: taskId,
        include_result: true
      });
      
      // Update our state
      await this.updateTaskState(taskId, response);
      
      // Stop tracking if complete
      if (['completed', 'failed', 'canceled'].includes(response.status)) {
        this.pendingTasks.delete(taskId);
      }
      
    } catch (error) {
      console.error(`Polling failed for task ${taskId}:`, error);
      
      // Exponential backoff on polling errors
      const task = this.pendingTasks.get(taskId);
      task.pollAttempts++;
      
      if (task.pollAttempts > 10) {
        console.error(`Giving up on task ${taskId} after 10 failed polls`);
        this.pendingTasks.delete(taskId);
      }
    }
  }
}

Webhook Event Format

AdCP webhook events include all necessary information for processing, with task-specific data in result:

{
  "operation_id": "op_123",
  "timestamp": "2025-01-22T10:25:00Z",
  "task_id": "task_456",
  "task_type": "create_media_buy",
  "status": "completed",
  "message": "Media buy created successfully",
  "result": {
    "media_buy_id": "mb_987654321",
    "packages": []
  }
}

Security Considerations

Webhook Authentication

Verify webhook authenticity using the authentication method specified during webhook registration:

function verifyWebhook(req, secret) {
  const signature = req.headers['x-adcp-signature'];
  const payload = JSON.stringify(req.body);
  const expectedSignature = createHmac('sha256', secret)
    .update(payload)
    .digest('hex');
  
  return signature === `sha256=${expectedSignature}`;
}

app.post('/webhooks/adcp', (req, res) => {
  if (!verifyWebhook(req, process.env.WEBHOOK_SECRET)) {
    return res.status(401).json({ error: 'Invalid signature' });
  }
  
  // Process webhook...
});

Replay Attack Prevention

Use timestamps and event IDs to prevent replay attacks:

function isReplayAttack(timestamp, eventId) {
  const eventTime = new Date(timestamp);
  const now = new Date();
  const fiveMinutes = 5 * 60 * 1000;
  
  // Reject events older than 5 minutes
  if (now - eventTime > fiveMinutes) {
    console.warn(`Rejecting old webhook event ${eventId}`);
    return true;
  }
  
  // Check if we've seen this event ID before
  return db.hasSeenWebhookEvent(eventId);
}

Best Practices Summary

Always implement polling backup - Don’t rely solely on webhooks
Handle duplicates gracefully - Use idempotent processing with event IDs
Check timestamps - Ignore out-of-order events based on timestamps
Return 200 quickly - Acknowledge webhook receipt immediately
Verify authenticity - Always validate webhook signatures
Log webhook events - Keep audit trail for debugging
Set reasonable timeouts - Don’t wait forever for webhook delivery
Graceful degradation - Fall back to polling if webhooks consistently fail

This reliability pattern ensures your application remains responsive and consistent even when webhook delivery is unreliable or fails entirely.

Reporting Webhooks

In addition to task status webhooks, AdCP supports reporting webhooks for automated delivery performance notifications. These webhooks are configured during media buy creation and follow a scheduled delivery pattern.

Configuration

Reporting webhooks are configured via the reporting_webhook parameter in create_media_buy:

{
  "buyer_ref": "campaign_2024",
  "reporting_webhook": {
    "url": "https://buyer.example.com/webhooks/reporting",
    "auth_type": "bearer",
    "auth_token": "secret_token",
    "reporting_frequency": "daily"
  }
}

Publisher Commitment

When a reporting webhook is configured, publishers commit to sending: (campaign_duration / reporting_frequency) + 1 notifications

One per frequency period during the campaign
One final notification at campaign completion
Delayed notifications if data isn’t ready within expected delay window

Payload Structure

Reporting webhooks deliver the same payload as get_media_buy_delivery with additional metadata:

{
  "notification_type": "scheduled",
  "sequence_number": 5,
  "next_expected_at": "2024-02-06T08:00:00Z",
  "reporting_period": {
    "start": "2024-02-05T00:00:00Z",
    "end": "2024-02-05T23:59:59Z"
  },
  "currency": "USD",
  "media_buy_deliveries": [
    {
      "media_buy_id": "mb_001",
      "buyer_ref": "campaign_a",
      "status": "active",
      "totals": {...},
      "by_package": [...]
    }
  ]
}

Notification Types:

scheduled: Regular periodic update
final: Campaign completed
delayed: Data not yet available (prevents missed notification detection)

Webhook Aggregation

Publishers SHOULD aggregate webhooks when multiple media buys share the same webhook URL, reporting frequency, and reporting period. This reduces webhook volume significantly for buyers with many active campaigns. Example: Buyer with 100 active campaigns receives:

Without aggregation: 100 webhooks per reporting period
With aggregation: 1 webhook containing all 100 campaigns in media_buy_deliveries array

Buyers must always handle media_buy_deliveries as an array, even when it contains a single media buy.

Timezone Handling

For daily and monthly frequencies, the publisher’s reporting timezone (from product’s reporting_capabilities.timezone) determines period boundaries:

Daily: Midnight to midnight in publisher’s timezone
Monthly: 1st to last day of month in publisher’s timezone
Hourly: UTC unless specified

Critical: Store publisher’s timezone when setting up webhooks to correctly interpret reporting periods.

Implementation Requirements

Array Handling: Always process media_buy_deliveries as an array (may contain 1 to N media buys)
Idempotent Processing: Same as task webhooks - handle duplicates safely
Sequence Tracking: Use sequence_number to detect gaps or out-of-order delivery
Fallback Strategy: Continue polling get_media_buy_delivery as backup
Delay Handling: Treat "delayed" notifications as normal, not errors
Frequency Validation: Ensure requested frequency is in product’s available_reporting_frequencies
Metrics Validation: Ensure requested metrics are in product’s available_metrics

See Optimization & Reporting for complete implementation guidance.

Error Handling

Error Categories

Protocol Errors - Transport/connection issues
- Handle with retries and fallback
- Not related to AdCP business logic
Task Errors - Business logic failures
- Returned as status: "failed" with error details
- Should be displayed to user
Validation Errors - Malformed requests
- Fix request format and retry
- Usually development-time issues

Error Response Format

Failed operations return status failed with details:

{
  "status": "failed",
  "message": "Unable to create media buy: Insufficient inventory available for your targeting criteria",
  "context_id": "ctx-123",
  "data": {
    "error_code": "insufficient_inventory",
    "requested_impressions": 10000000,
    "available_impressions": 2500000,
    "suggestions": [
      "Expand geographic targeting",
      "Increase CPM bid",
      "Adjust date range"
    ]
  }
}

Human-in-the-Loop Workflows

Design Principles

Optional by default - Approvals are configured per implementation
Clear messaging - Users understand what they’re approving
Timeout gracefully - Don’t block forever on human input
Audit trail - Track who approved what when

Approval Patterns

async function handleApprovalWorkflow(response) {
  if (response.status === 'input-required' && needsApproval(response)) {
    // Show approval UI with context
    const approval = await showApprovalUI({
      title: "Campaign Approval Required",
      message: response.message,
      details: response.data,
      approver: getCurrentUser()
    });
    
    // Send approval decision
    const decision = {
      approved: approval.approved,
      notes: approval.notes,
      approver_id: approval.approver_id,
      timestamp: new Date().toISOString()
    };
    
    return sendFollowUp(response.context_id, decision);
  }
}

Protocol-Agnostic Examples

Product Discovery with Clarification

// Works with both MCP and A2A
async function discoverProducts(brief) {
  let response = await adcp.send({
    task: 'get_products',
    brief: brief
  });
  
  // Handle clarification loop
  while (response.status === 'input-required') {
    const moreInfo = await promptUser(response.message);
    response = await adcp.send({
      context_id: response.context_id,
      additional_info: moreInfo
    });
  }
  
  if (response.status === 'completed') {
    return response.data.products;
  } else if (response.status === 'failed') {
    throw new Error(response.message);
  }
}

Campaign Creation with Approval

async function createCampaign(packages, budget) {
  let response = await adcp.send({
    task: 'create_media_buy',
    packages: packages,
    total_budget: budget
  });
  
  // Handle approval if needed
  if (response.status === 'input-required') {
    const approved = await getApproval(response.message);
    if (!approved) {
      throw new Error('Campaign creation not approved');
    }
    
    response = await adcp.send({
      context_id: response.context_id,
      approved: true
    });
  }
  
  // Handle async creation
  if (response.status === 'working') {
    response = await waitForCompletion(response);
  }
  
  if (response.status === 'completed') {
    return response.data.media_buy_id;
  } else {
    throw new Error(response.message);
  }
}

Migration Guide

From Custom Status Fields

If you’re using custom status handling: Before:

{
  "clarification_needed": true,
  "approval_required": true, 
  "processing": false
}

After:

{
  "status": "input-required",
  "message": "Budget exceeds limit. Please approve to proceed."
}

Backwards Compatibility

During the transition period, responses may include both old and new fields:

function getStatus(response) {
  // New unified approach
  if (response.status) {
    return response.status;
  }
  
  // Backwards compatibility
  if (response.clarification_needed) return 'input-required';
  if (response.approval_required) return 'input-required'; 
  if (response.processing) return 'working';
  
  return 'completed'; // Default assumption
}

Next Steps

MCP Integration: See MCP Guide for tool calls and context management
A2A Integration: See A2A Guide for artifacts and streaming
Protocol Comparison: See Protocol Comparison for choosing between MCP and A2A

This unified status approach ensures consistent behavior across all AdCP implementations while making client development more predictable and robust.

Getting Started

Protocols

adagents.json

Signals Protocol

Curation Protocol

Media Buy Protocol

Creative

Reference

Community

​Task Status System

​Status Values

​Response Structure

​Client Decision Logic

​Basic Status Handling

​Advanced Status Patterns

​1. Clarification Flow

​2. Approval Flow

​3. Long-Running Operations

​Async Operations

​Operation Types

​Timeout Handling

​Context Management

​Session Continuity

​Context Expiration

​Task Management & Webhooks

​Task Tracking

​Push Notification Architecture

​MCP Webhooks

​A2A Webhooks

​Unified Data Schemas

​When Webhooks Are Called

​Push Notification Format by Protocol

​MCP: HTTP Webhook POST

​A2A Webhook POST

​Status-Specific Data Schemas

​Supported Async Operations

​MCP Webhook URL Patterns

​Task State Reconciliation

​Status Progression

​Webhook Reliability

​Delivery Semantics

​Security

​Webhook Authentication (Required)

​Retry and Circuit Breaker Patterns

​Retry Strategy

​Circuit Breaker Pattern

​Queue Management

​Implementation Requirements

​Idempotent Webhook Handlers

​Sequence Handling

​Polling as Backup

​Webhook Event Format

​Security Considerations

​Webhook Authentication

​Replay Attack Prevention

​Best Practices Summary

​Reporting Webhooks

​Configuration

​Publisher Commitment

​Payload Structure

​Webhook Aggregation

​Timezone Handling

​Implementation Requirements

​Error Handling

​Error Categories

​Error Response Format

​Human-in-the-Loop Workflows

​Design Principles

​Approval Patterns

​Protocol-Agnostic Examples

​Product Discovery with Clarification

​Campaign Creation with Approval

​Migration Guide

​From Custom Status Fields

​Backwards Compatibility

​Next Steps

Task Status System

Status Values

Response Structure

Client Decision Logic

Basic Status Handling

Advanced Status Patterns

1. Clarification Flow

2. Approval Flow

3. Long-Running Operations

Async Operations

Operation Types

Timeout Handling

Context Management

Session Continuity

Context Expiration

Task Management & Webhooks

Task Tracking

Push Notification Architecture

MCP Webhooks

A2A Webhooks

Unified Data Schemas

When Webhooks Are Called

Push Notification Format by Protocol

MCP: HTTP Webhook POST

A2A Webhook POST

Status-Specific Data Schemas

Supported Async Operations

MCP Webhook URL Patterns

Task State Reconciliation

Status Progression

Webhook Reliability

Delivery Semantics

Security

Webhook Authentication (Required)

Retry and Circuit Breaker Patterns

Retry Strategy

Circuit Breaker Pattern

Queue Management

Implementation Requirements

Idempotent Webhook Handlers

Sequence Handling

Polling as Backup

Webhook Event Format

Security Considerations

Webhook Authentication

Replay Attack Prevention

Best Practices Summary

Reporting Webhooks

Configuration

Publisher Commitment

Payload Structure

Webhook Aggregation

Timezone Handling

Implementation Requirements

Error Handling

Error Categories

Error Response Format

Human-in-the-Loop Workflows

Design Principles

Approval Patterns

Protocol-Agnostic Examples

Product Discovery with Clarification

Campaign Creation with Approval

Migration Guide

From Custom Status Fields

Backwards Compatibility

Next Steps