Threll.ai
For developers

Build Threll into the systems you already run

One API call starts a conversation. Webhooks let your systems join in while it happens. And with MCP, your AI assistant can do the work for you. Here are the essentials – the documentation has the rest.

Open the documentation (opens in a new window)Book Meeting (opens in a new window)
Your existing workflow

Fits in where you already work

You don't need to build anything new around Threll. Your workflow says when to call, and the outcome lands where it belongs.

  1. 1

    One API key

    Create a key under Settings → Developer and send it in the x-api-key header. That's all authentication takes.

  2. 2

    One request starts the call

    Say who to call, which agent calls and what the goal is. The call is queued right away, or scheduled for a set time.

  3. 3

    Events straight to your endpoint

    Status, transcript and the final signal stream to a webhook you own while the call is in progress.

  4. 4

    Results back in your system

    Fetch the analysis, next action, transcript and recording over REST once post-processing is done.

Start an outbound call · curl
curl -X POST https://api.threll.io/v1/accounts/$ACCOUNT_ID/phone-calls \
  -H "x-api-key: $THRELL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "workerId": "2b7d9a4c-6e3f-48a1-9c5d-8e0b4f2a1c8d",
    "phoneNumber": "+4799999999",
    "context": "Qualify this demo request. Goal: book a meeting.",
    "customer": {
      "type": "lead",
      "firstName": "Kari",
      "externalId": "crm-8841"
    }
  }'

What a call looks like from the outside

  1. Synccall.worker_request

    Before the agent answers: give it fresh data about the caller.

  2. Asynccall.status_update

    The call is ringing, in progress or finished.

  3. Asynccall.transcript

    Each turn is sent as soon as it's been spoken.

  4. Synccall.tool_call

    The agent needs something from your system. The call waits for your answer.

  5. Asynccall.ended

    The final event. Now you can fetch the results.

Business systems

Connect your business systems

CRM, patient records, booking or ERP: the agent reads and writes data in your system while it talks to the customer. You decide what it can reach.

  1. Before the call

    Give the agent context

    Look the caller up in your CRM and send along their history, open cases or past appointments. The agent then answers as if it knows the customer.

    call.worker_requestSync
  2. During the call

    Let the agent do the work

    Check free slots, look up an order or log a case: the agent calls the tools you define, and you answer from your system.

    call.tool_callSync
  3. After the call

    Sync the outcome

    Update the customer record, create missing contacts and route the next action – like "book meeting" or "escalate to a human" – to your ticketing system.

    call.endedAsync
One endpoint for every event · Node
app.post('/threll/webhooks', async (req, res) => {
  const { type, data } = req.body;

  switch (type) {
    case 'call.worker_request': {
      const customer = await crm.findByPhone(data.customer.number);
      return res.json({ context: customer?.summary ?? 'New caller' });
    }

    case 'call.tool_call': {
      const account = await erp.findAccount(data.arguments.accountNumber);
      return res.json({ result: account ?? { error: 'not found' } });
    }

    case 'call.ended':
      queue.push({ syncCall: data.callId });
      return res.sendStatus(200);

    default:
      return res.sendStatus(200);
  }
});

Safe to build on

  • Every webhook is signed with HMAC-SHA256 in X-Threll-Signature, so you know it came from Threll.
  • X-Threll-Event-Id makes it easy to ignore events that are delivered again.
  • Sync events wait for your reply, so keep lookups under a second – the customer is on the line.

Ready-made integrations

  • Cal.com
  • EasyPractice
  • Magento
  • Intercom
  • Crisp
  • GoHighLevel
  • Gastroplanner
  • Shipmondo
  • Vipps

Using something else? Connect it yourself with webhooks and your own tools – no need to wait for us.

MCP

Use Threll from your AI assistant

Threll has an MCP server, so Claude and other MCP clients can run your agents, schedule calls and search call history – with exactly the access the user approves.

MCP serverhttps://mcp.threll.io

Set up in a couple of minutes

  1. 1

    Add a custom connector

    Paste the server address into your client. The rest of the configuration is discovered automatically.

  2. 2

    Sign in and approve access

    Sign-in uses OAuth 2.1. There are no API keys to store, and access can be revoked at any time.

  3. 3

    The tools are ready

    The assistant only gets the tools the user granted – account administration, for example, or a specific integration.

A human confirms

  • Calls are prepared first and only placed once someone confirms.
  • Up to 100 calls can be prepared and confirmed together.
  • Scheduled calls can be cancelled before they go out.
Example

Have Harald call these five leads tomorrow at nine.

Five calls from Harald are ready for tomorrow at 09:00. Shall I place them?

Clients without connector support · JSON
{
  "mcpServers": {
    "threll": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.threll.io/"]
    }
  }
}
Documentation

Everything you need to go further

Guides, references and examples live at developers.threll.ai.

Need help with your integration?

Book a meeting and we'll walk through the setup together – or ask a quick question in the chat.

Book a meeting (opens in a new window)

Questions from developers

Can't find the answer here? Ask us in the chat.

You create API keys in the platform under Settings → Developer. The key goes in the x-api-key header on every request.

No. Add the MCP server as a connector in your AI assistant and sign in. You approve the access yourself and can revoke it whenever you like.

Every request carries an X-Threll-Signature header: an HMAC-SHA256 of the raw body, signed with your subscription's secret. Verify it before you parse the body.

Define your own tools and answer them from your webhook endpoint. The agent can then read from and write to any system you have an API for. If you'd like a hand, book a meeting with us.