Build with

Overview

Voiceflow integrates with Zoom Contact Center to answer selected customer calls, transfer callers to staffed queues and give staff an on-demand assistant. Keep your existing customer number and Zoom routing; send the requests you choose to a Voiceflow phone agent.

The call recipe uses a Voiceflow phone number and a separate Zoom number for human support. Agent assist runs in your authenticated CRM or staff application, alongside the Zoom Contact Center softphone. Neither connection automatically copies Zoom engagement variables or transcripts into Voiceflow.

Examples of when to use Zoom Contact Center

  • Answer inbound calls — route an IVR branch to Voiceflow for delivery updates and policy questions.
  • Human agent hand-off — let callers reach a Zoom queue when they need a person.
  • Add agent assist — give staff policy guidance and reply drafts beside their Zoom controls.
  • Test and launch — verify the caller journey and rehearse routing rollback before changing the public entry point.

What you need

  • A Zoom Contact Center license and an administrator who can edit voice Flows, Queues and Entry Points. Zoom Phone call handling uses a different setup.
  • A Voiceflow phone project, an editable test Environment and a connected phone number.
  • A dedicated Zoom human-support number that bypasses the AI branch and reaches the receiving queue.
  • For Code, cURL, jq and the current Voiceflow CLI with Agent v2 support. Set PROJECT_ID, ENVIRONMENT_ALIAS and a server-side personal or service key in VF_TOKEN. Configure CLI authentication with the same key, or pass --token "$VF_TOKEN". These credentials do not authorize Zoom administration.

Telephone routing, SIP bots and chat bots

Choose the connection contract before configuring a Zoom Bot widget.

Zoom’s voicebot framework requires coordinated SIP provisioning, TLS/SRTP and delivery of context before SIP REFER. Follow its SBC requirements and data mapping when developing that adapter. Do not substitute the phone-number recipe for a Bot widget’s SIP endpoint.

For chat, Zoom’s action/trigger guide and messaging protocol define a separate relay contract. It must preserve opaque customData, bind Zoom and Voiceflow sessions, transform messages and handle human transfer and conversation closure. A Voiceflow runtime URL alone cannot be entered as that relay. These adapters are not included in the telephone or staff-assist examples below.

Preparing your voice agent

Resolve the supported customer requests before transferring callers to Zoom Contact Center staff.

Configure the conversation

Use short spoken replies, ask one question at a time and state when the agent should bring in a person. Attach the Playbooks, Knowledge and tools needed for the requests you actually support. Connecting a Zoom Contact Center number does not import company policies or create business-system tools.

Studio

1

Prepare the phone agent

Open the project in Agent studio, select the test Environment, then open Agent → Instructions. Define the support tasks, approved policy sources and transfer conditions. Attach the relevant Playbooks and tools, then insert their resource references in the instructions.

Connect the agent's instructions to the Playbooks, Knowledge and tools that resolve the caller's request. Photo:

The example retailer uses its own order lookup service. It is not a Zoom Contact Center customer-record API.

2

Check speech and escalation

Configure the agent’s Voice settings. Test a policy question, an order question and a request for a person. Configure the actual transfer destination in Human agent hand-off below before testing a phone transfer. Run the agent after edits to compile its Draft before calling the connected line.

Code

Start with an existing phone project and attached Playbooks/tools. Save your Instructions as agent.json; use the names of resources already attached to that agent. These fields leave its existing resource registries and Voice settings unchanged:

agent.json
Update the test agent

Read the agent back with Get agent. Confirm the intended Instructions and tools remain configured. Compile the Draft after edits using Testing and launch below; a saved configuration is not yet proof of a working telephone conversation.

Routing calls to Voiceflow

Keep your Zoom Contact Center number and send a selected call-flow branch to the AI agent.

Assign the Voiceflow destination

Use a dedicated test line first. Your Zoom Contact Center administrator needs its full E.164 number and a reachable human fallback before editing the Zoom voice Flow.

Studio

1

Find the AI entry number

In Agent studio, open your phone agent’s project. In the left sidebar, open Phone numbers and find the connected test line. Copy its full number, including the country code.

Use this Voiceflow line as the external destination of the Zoom Contact Center test branch. Photo:
2

Select the test version

Open the line’s Environment assignment, choose the test Environment and Draft, then save. Call this number directly and confirm the intended agent answers before testing through Zoom Contact Center.

Match the line's Environment and version to the agent you are testing. Photo:

Code

Use a personal or service key with access to this workspace and phone project. Set WORKSPACE_ID to the workspace containing the connected line. Obtain it through Workspaces, and keep PROJECT_ID and ENVIRONMENT_ALIAS from the agent configuration above. The key needs vf.phone-number:LIST and vf.phone-number:READ for the workspace, plus vf.phone-number:ASSIGN on the target Environment. The number and project must belong to the same workspace.

1

Find the connected line

List the workspace

Select the dedicated test line from phoneNumbers. Set PHONE_NUMBER_ID to its id, not its E.164 value. Keep value for the Zoom Contact Center administrator and retain its current voice assignment for rollback.

2

Pin the voice side to the test Draft

Create assignment.json and assign the line

Verify phoneNumber.voice.projectID, environmentAlias and version are the intended project, test alias and draft. Assignment changes the voice side only; it leaves SMS routing unchanged.

Compile the Draft after agent edits, then call the Voiceflow number directly. Confirm the expected greeting and two-way speech before giving the number to the Zoom Contact Center administrator. The Zoom Contact Center flow changes below use Zoom Contact Center administration, not the Voiceflow credential.

Configure the Zoom voice Flow

Use a dedicated Zoom test number and a draft voice Flow. Record the previous entry-point assignment and routing before editing. Your Zoom administrator needs the tested Voiceflow phone number and the receiving human queue.

1

Open the pilot Flow

In the Zoom web portal, open Profile → Admin Center → CX management → Contact Center → Flows. Select the pilot voice Flow. Use the Start widget’s Manage Entry Point control to confirm the test number reaches this Flow. See Start widget settings.

2

Add the Voiceflow phone destination

Drag Route To onto the intended AI branch. Select it, open Settings and choose OutgoingCall. Set Outgoing Call Target → Phone Number and Outgoing Number Type → Number. Select the number’s country/region and enter the dialable number without spaces or dashes. Use the line’s phone-number value, not its Voiceflow resource ID.

3

Connect the route's exits

Set Max Ring Time and Caller ID deliberately. Connect Call Failed and Call Not Answered to the pilot’s human fallback or announcement, and configure Call Answered for the successful route. Zoom’s Route To settings describe these exits. Queue overflow is a separate setting; do not assume the outgoing-call route uses a queue’s exits.

4

Publish only the pilot and place a call

Save and publish the pilot Flow, confirm its test entry-point assignment, then call that number. Choose the AI branch and confirm the expected agent answers with two-way speech. Exercise an unreachable destination and the human fallback before moving customer traffic.

Keep customer context explicit

  • Engagement data: this telephone route does not map Zoom Flow variables, engagement IDs or transcripts into the Voiceflow session.
  • Identity: caller ID is not account authentication. Use your own trusted service to verify requests before returning protected customer data.
  • Continuation: use the independent human-support number below to reach Zoom staff. Do not assume an external phone transfer resumes a Bot widget.

Human agent hand-off

Transfer callers to a separate Zoom Contact Center number that reaches a staffed queue.

Prepare the Zoom return route

Give the human-support number a separate Zoom voice Flow. In Admin Center → CX management → Contact Center → Flows, connect its Start widget to Route To → Queue and select the receiving queue. Confirm its Manage Entry Point assignment uses the human number, not the number that routes to AI. Test staffing, queue overflow and closed-hours handling with the receiving team.

Configure Call forward

Use Call forward when the caller requests a person or needs an action outside the AI agent’s supported tasks. Set the destination to the dedicated Zoom Contact Center human-support number, not the public number that routes to AI.

Studio

1

Open the phone agent's transfer tool

In Agent studio, select the test Environment and open Agent → Call forward. Enable the tool and choose Phone number.

2

Set the human destination

Enter the dedicated Zoom Contact Center return number with its country code. Describe the requests that require a person, then close the editor to return to the agent.

Forward to the human-only Zoom Contact Center number, never to the public number's AI branch. Photo:
3

Test with an answered call

Call through the Zoom Contact Center test entry, ask the AI for a person and have a teammate answer the queue. Confirm two-way audio and that the AI stops speaking. Use a connected telephone line: prototype test calls cannot forward.

Code

Save this configuration as transfer.json, replacing the illustrative number with your actual human-only Zoom Contact Center number. Updating the tool replaces its settings; preserve existing extension, message or caller-ID options if you use them.

transfer.json
Set the human destination

Read the agent back, confirm the return number, compile the test Draft and place a real call through Zoom Contact Center. A successful PATCH or CLI exit does not prove the receiving queue answered.

Understand what crosses the transfer

Test by calling through the Zoom Contact Center test number and asking the AI for a person. Have a teammate answer the human queue; confirm two-way audio and that the AI stops speaking.

  • Queue selection: Zoom Contact Center’s receiving flow chooses the queue. Call forward dials a phone number, not a Zoom Contact Center queue ID.
  • Customer context: this phone transfer does not deliver an automatic summary, transcript or Zoom engagement variables. Have the human confirm the case reference, or implement a separately authorized business-system context exchange.
  • Failures: a rejected forward can end the call; this path does not expose an automatic Voiceflow busy/no-answer recovery branch. Test the destination’s closed-hours and unavailable-agent behavior.

Adding agent assist

Give staff policy guidance and reply drafts beside Zoom in their CRM workspace.

This is an on-demand assistant, separate from the caller’s phone session. Staff ask a question and review the suggestion before speaking or sending it to the customer. The recipe does not listen to calls, transcribe audio or send customer replies automatically.

Prepare the staff assistant

Studio

Open a separate staff-assistant project in Agent studio → Agent → Instructions. Tell it to answer the staff member, use approved Knowledge and label suggested customer replies. Keep the assistant read-only; the application below does not propagate staff permissions into agent tool calls.

Keep suggestions staff-facing and require human review before replying to the customer. Photo:

Publish to a dedicated test Environment. Give your integration developer the project ID and Environment alias, and provision a server credential through your team’s secret-management process. Obtain the authenticated application URL and approved CRM deployment before continuing below.

Code

Use a ready read-only staff assistant and its Published test Environment. Download assist.mjs and client.mjs. Mount the handler as a same-origin POST /api/agent-assist route in your Node.js 22+ staff application:

Authorize the staff member and selected case

Implement requireStaff(request) with the application’s verified staff session. Implement loadCaseForStaff(staff, caseId) against your trusted case service: return {id: caseId, context: approvedText} only if that staff member can read the case; otherwise return null. Set the allowed fields on the server. A Zoom sign-in or browser-supplied case ID is not authorization to your case service. Apply your host’s CSRF protection for cookie sessions.

The handler starts an isolated Voiceflow session, sends launch, then streams the staff question with the authorized context. Voiceflow credentials and runtime session keys stay on the server. Build a case picker from the staff member’s authorized cases; a Zoom telephone transfer does not populate it automatically.

Request a suggestion from the embedded application

casePicker, questionInput, askButton and output are elements in your application’s UI. Populate the picker through an authenticated case-list endpoint; reauthorize each request in the handler. Clear its selection and any pending answer when staff finish or change the case. Render suggestions with textContent, not innerHTML.

Put Zoom and the assistant in the staff workspace

Zoom Smart Embed embeds the Zoom Contact Center softphone inside your web application. It does not embed an arbitrary assistant into the Zoom desktop app. Your developer hosts the assistant beside that softphone in the CRM; your Zoom administrator configures and approves the Smart Embed integration for that application.

1

Prepare the authenticated CRM application

The integration developer deploys the staff-only app on HTTPS, mounts the authorized assistance endpoint and provides a case picker populated from your trusted case service. Give the Zoom administrator the exact CRM origin and deployment URL. Staff must be able to sign in and read their permitted cases before the assistant is enabled.

2

Add Zoom Smart Embed

Follow Zoom’s Smart Embed setup to configure the integration and embed its softphone iframe in that CRM. Set the iframe’s origin query parameter to the CRM’s actual HTTPS origin. Use Zoom’s documented sandbox and media permissions; complete Zoom sign-in inside the embedded softphone. If your account uses SAML SSO, use its documented vanity-domain URL.

Studio

Ask your integration developer to initialize Smart Embed in the CRM, keep the Voiceflow API key on the server and provide the authorized case selector. Confirm both the Zoom login and CRM login work with a pilot staff account before requesting a suggestion.

Code

Download softphone.mjs. It sets Zoom’s documented iframe permissions and handles the initialization handshake, checking the exact origin and source window. Add the iframe beside your existing assistant UI:

CRM softphone host

Use the approved vanity origin instead of https://zoom.us when your SAML setup requires it. Permit that exact frame origin in the CRM’s frame-src policy. The sample initializes the softphone only; keep the assistance route, authorized case picker and case-change cancellation code above. Do not turn Zoom events into automatic case access.

3

Test the staff workflow

Sign in as an assigned staff member and take a pilot call through the embedded Zoom softphone. Select an authorized case in the CRM, ask for guidance and confirm the reply stays staff-facing. Change the selected case while a response is streaming: the old request must abort and its output must clear. Have staff review a draft before sending or speaking it.

Keep assistance separate from call control

  • Data selection: this example uses an authorized CRM case, not an automatic Zoom transcript feed. A telephone route does not provide a Zoom engagement-to-case mapping.
  • Read-only scope: the assistant suggests answers; it does not write Zoom notes, transfer the live call or submit a customer reply.
  • Browser messages: if you add Smart Embed events later, check both the exact Zoom origin and the iframe’s contentWindow before handling a message. Do not use an event payload as staff or case authorization.

Testing and launching

Prove the telephone and staff-assistance paths before routing customer traffic.

Compile and call the test Draft

Studio

After changing the agent, run a test from Agent → Run. Keep the isolated phone line assigned to the test Environment’s Draft, then place a real call through the Zoom Contact Center test number. Studio’s test conversation alone cannot verify external telephone routing.

Code

Compile after agent or tool edits. The line assigned earlier must target this same Environment’s Draft:

Compile the test Draft

Confirm the compile succeeds, then call the Zoom Contact Center test number. This is the phone path; creating a text runtime session is not an equivalent test.

Exercise the complete journey

  • Self-service: a policy question and a real test-order question reach the intended Playbook and produce spoken replies.
  • Human agent hand-off: a request for a person reaches the human-only Zoom Contact Center number, a teammate answers and both parties can hear each other. Confirm no route loops back to the AI branch.
  • Unavailable destination: exercise the Zoom Contact Center route’s unsuccessful-transfer handling and the human queue’s closed-hours handling separately.
  • Caller exit: hang up during the AI conversation; verify the telephone leg ends. Check that the external destination does not leave a connected call leg behind.
  • Agent assist: confirm sign-in, allowed/denied case access and clearing a pending answer when changing case inside the actual authenticated CRM workspace.

Publish and retain rollback

Studio

In the test Environment, choose Publish, enter a release name and publish the tested configuration. Open Phone numbers, select the connected line’s Environment assignment, change Draft to Published and save. Keep the intended Environment selected.

Code

Publish only after the Draft telephone checks pass. This creates a release and updates that Environment’s Published version:

Publish the tested Environment

Check the returned release/Environment, then read the phone line back using the API or CLI get command above. Confirm phoneNumber.voice.version is published with the intended project and Environment alias. Publishing alone does not change a Draft line’s assignment. The API procedure reuses assignment.json from Routing calls to Voiceflow.

Call the Published line again before changing the public Zoom Contact Center number to the approved Flow. Retain the previous Zoom Flow and entry-point assignment and the prior Voiceflow release so the administrator can restore routing if the live test fails.

Use the shared Test and launch your integration guide for the release handover.

Use the up and down arrow keys to select a result, Enter to open it, and Escape to close the search.