ClawCall vs Vapi

What does it take to place one outbound call on Vapi?

Vapi's phone quickstart has three steps before the phone rings. First you create an assistant, which holds the model provider, the model, the system prompt and the voice. Then you create a phone number attached to that assistant, either a free Vapi number in a US area code or a number imported from another provider. Then you create the call with the assistant ID, the phone number ID and the customer's number.

Each piece is reusable, which is the point: a team building a product sets up assistants once and places thousands of calls. Vapi's create-call endpoint also accepts a transient assistant configuration inside the request, so you can skip saving an assistant, but you still send the full model and voice settings and a phone number ID. ClawCall collapses all of this into one request.

The model, voice, telephony and phone number are already chosen, so the request carries only the destination and the task.

# 1. Create an assistant (model, voice, prompt)
curl -X POST https://api.vapi.ai/assistant \
  -H "Authorization: Bearer $VAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Store caller",
    "model": {
      "provider": "openai",
      "model": "gpt-4o",
      "messages": [{ "role": "system", "content": "Ask whether the store has 10mm hex keys in stock and when it closes today." }]
    },
    "voice": { "provider": "11labs", "voiceId": "your-voice-id" }
  }'

# 2. Create a phone number for the assistant
curl -X POST https://api.vapi.ai/phone-number \
  -H "Authorization: Bearer $VAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "provider": "vapi", "assistantId": "your-assistant-id", "numberDesiredAreaCode": "415" }'

# 3. Place the call
curl -X POST https://api.vapi.ai/call \
  -H "Authorization: Bearer $VAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "assistant": { "assistantId": "your-assistant-id" },
    "phoneNumberId": "your-phone-number-id",
    "customer": { "number": "+15551234567" }
  }'

curl -X POST https://api.clawcall.dev/call \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: $CLAWCALL_API_KEY" \
  -d '{
    "to": "+15551234567",
    "task": "Ask whether the store has 10mm hex keys in stock and when it closes today."
  }'

# Response
{ "call_id": "ba645d75-...", "status": "queued" }

After the call is placed, Vapi reports progress through webhooks to the server URL you configure, and the call record is available from its API. ClawCall is fire-and-poll: poll GET /call/{call_id} every three seconds until lifecycle is finalized, then read the outcome, talk_seconds and captured transcript evidence when present. MCP hosts never see this loop; the ClawCall connector's tools place the call and check its status for them.

What does a 40-minute hold cost on Vapi, Bland and ClawCall?

Vapi's billing guide says every minute an assistant runs is billed, and a call that sits in a hold queue keeps the assistant running. On usage-only billing, a 40-minute call costs $2.00 in Vapi hosting fees alone. The sample providers in Vapi's own calculator (Deepgram for transcription, OpenAI for the model and ElevenLabs for the voice) add $0.0318 to $0.0789 a minute, which brings the call to $3.27 to $5.16 before telephony.

Vapi also ends calls after 600 seconds unless you raise maxDurationSeconds on the assistant, so a long hold needs that setting changed first.

On ClawCall, the same call costs nothing beyond the monthly plan, because paid plans include fair-use unlimited outbound calls with no per-minute charges. The math favors Vapi in other cases: short, high-volume calls where you want to choose cheaper providers, or a product where per-call cost is passed on to your own customers.

Build the agent or hand off the call

With Vapi, your team owns the behavior of every call. You write the system prompt, decide when the assistant presses keys, configure voicemail handling with Vapi's voicemail tool, set silence and duration limits, and decide whether and how the assistant says it is an AI. That control is valuable when the calls are your product, for example a clinic's appointment line or a lender's reminder campaign, and it is why Vapi offers squads, simulations, observability and testing tools.

With ClawCall, those decisions are already made and documented. The agent navigates phone menus, waits on hold, talks to the person who answers, always says it is an AI when asked, leaves a voicemail only when your instructions authorize one, and does not send bulk or robocall-style promotions, while individual outreach calls are allowed. When the call needs you, for example to confirm your identity or approve a charge, loop_in_user calls your verified phone and asks whether you want to join before connecting you. You trade configurability for a call that works the same way every time.

Which one fits an AI coding agent or a chat assistant?

Both can be driven by an AI agent through their REST APIs, so the question is how much the agent has to manage. A Vapi integration needs a Vapi API key, a saved or transient assistant, a phone number, and webhook or polling handling, and the agent's operator pays for every provider involved. ClawCall's preferred path for agent hosts is its hosted OAuth MCP connector at https://api.clawcall.dev/mcp.

ChatGPT, Claude and other hosts that support remote MCP sign in with OAuth, so no API key is pasted into a chat, and the connector exposes twelve bounded tools for placing calls, checking status, reading what was captured and changing call settings. Hosts without remote MCP use the packaged skill for Claude Code, Cursor, ClawHub and OpenClaw, or the REST API.

ClawCall's limits matter here too: it calls US +1 numbers only, in English, and each account places one outbound call at a time on every plan. If your agent needs to phone other countries, speak Spanish or run hundreds of calls at once, Vapi or another platform is the right base.