Using the MCP

Prompt Cookbook

Copy-paste prompts for everything the MCP can do. Just type them in your editor’s chat.

View as Markdown

No commands to memorize. Just ask in plain English — the AI picks the right tool.

Agents

List all my agents
Show me agents with "support" in the name
Get full details for agent "Sales Outreach" — voice, model, prompt, and all settings
What prompt is "Support Bot" using?
Create a new agent called "Customer Support"
Create an agent called "Survey Bot" with Hindi as the default language and yuvika voice
Create an agent called "Appointment Scheduler" with gpt-4.1 model,
set its prompt to "You schedule appointments for a dental clinic...",
and set the first message to "Hello! I'm calling from SmileCare Dental."
Update the prompt for "Support Bot" to:
"You are a helpful customer support agent for Acme Corp.
Be polite, concise, and always confirm before taking action."
Add this to the beginning of "Sales Bot" prompt:
"Always introduce yourself by name before anything else."
Change the voice of "Support Bot" to yuvika
Change the first message of "Support Bot" to "Hi, thanks for calling Acme support! How can I help?"
Update "Support Bot" to support English and Hindi with language switching enabled
Switch "Sales Bot" to electron model
Archive the agent called "Old Test Bot"
Unarchive "Old Test Bot"
Duplicate "Sales Bot" so I can experiment without touching the original

When an agent has versioning enabled, config changes create a draft that you publish explicitly:

Change the voice of "Production Agent" to rachel

AI responds: “Config saved as draft. Want me to publish it?”

Yes, publish it with the label "Switched to rachel voice"
Discard the draft — I changed my mind

Edit tools accept a draft_id, so a prompt rewrite, a pre-call API and a full set of API-call tools can land in a single draft and go live with one publish:

For "Support Agent": rewrite the prompt for inbound customer support, set the
pre-call API to https://api.acme.com/customers/lookup keyed on the caller's
number, and add tools for order lookup, refund status and address change —
all in ONE draft, then publish it with the label "Inbound support v2"

The AI stacks each edit onto the same draft and publishes once. Publishing waits for the automatic prompt security check, then activates the version and reports whether it’s live.

When adding several API-call tools to one draft, the AI passes them as a single batch — the draft’s tools section is replaced wholesale on each write, so the last write must contain the complete set.
Show me all active drafts for "Production Agent"
What changed in the latest draft compared to the live version?
Rename the draft to "Voice experiment - rachel"
Test the draft with a webcall before publishing
Test the draft with a phone call to +14155551234
Show me the version history for "Production Agent"
What's different between version 3 and the current live version?
Roll back "Production Agent" to version 2
Pin version 3 with the label "Best performing config"
Compare call metrics between version 2 and version 3 for last month
Test version 2 of "Production Agent" with a webcall
Test version 3 with a phone call to +14155551234

Making Calls

These trigger real phone calls. The agent will call the number and follow its configured prompt.

Call +14155551234 using the "Sales Outreach" agent
Call +916366821717 using agent "Support Bot"
Call +14155551234 using "Sales Outreach" with variables:
prospect_name = "John", company = "Acme Corp"
What phone numbers do I have?

Then:

Call +14155551234 using "Sales Outreach" from the +18005551234 number
What's the status of call CALL-1234567890-abc123?

Testing Prompts (Text Chat)

chat_with_agent talks to a published agent in text — no audio, no phone — over the realtime chat WebSocket. It’s the fastest way to exercise a prompt and read back exactly what the agent says, and it’s the basis for an automated build → test → refine loop. (For an unpublished draft, start one with test_draft in chat mode.)

Chat with the "Hotel Booking" agent: say "Hi", then "Book a table for two tomorrow at 7pm"

Returns the full transcript — the greeting, your turns, and the agent’s replies.

Chat with agent "Sales Outreach" — open with "Hi, who is this?" — using
variables prospect_name = "John" and company = "Acme Corp"
Chat with the "Support Bot" agent and run a few tricky customer questions.
Read the transcript, tell me where the prompt falls short, update the prompt
to fix it, then run the same conversation again and compare.

The editor chains chat_with_agent → reviews the transcript → update_agent_prompt → re-runs — an automated prompt-tuning loop, with no test phone calls.

Debugging Calls

Call IDs look like CALL-1234567890-abc123. You can find them in call list responses or in the console.

Debug call CALL-1234567890-abc123

Returns everything: transcript, errors, latency metrics, cost breakdown, post-call analytics, variables, and agent config at time of call.

What was the transcript for CALL-1234567890-abc123?
Why did CALL-1234567890-abc123 fail?
What was the cost breakdown for that last call?

Browsing Call History

Show me all calls from the last 24 hours
Show me calls from this week
Show me failed calls from this week
Get all completed calls for agent "Support Bot"
Show me all inbound calls this month
List outbound calls from the last 7 days
Show webcalls from today
Find all calls to +14155551234
Show calls from campaign 682abc123def456

Text-to-Speech

Convert this to speech with yuvika voice:
"Welcome to Smallest AI. We build the fastest voice agents on the planet."
Generate an MP3 with the daniel voice at 1.5x speed:
"Your order has been confirmed and will arrive tomorrow."
Show me available female voices
List voices that support Hindi

Then:

Generate speech with that voice: "Namaste, aapka order confirm ho gaya hai"

Speech-to-Text

Transcribe this audio file: ~/Desktop/recording.wav

The AI will ask what language the audio is in before transcribing.

Transcribe the audio at https://example.com/meeting.mp3
The language is English.
Transcribe ~/Desktop/meeting.wav in English with speaker diarization and word timestamps
Transcribe this call recording with emotion detection: ~/Desktop/call.wav
Transcribe with PII redaction: ~/Desktop/support-call.mp3

Analytics

Show me the full dashboard for the last 30 days
Give me a dashboard for agent "Sales Bot" this month
What's our total call spend this week?
Show credit usage for the last 30 days
How many calls did we make in January?
Compare agent performance this month — who has the most calls?
Which agent has the highest completion rate?
What are the peak hours for calls this week?
Show concurrent call counts for today
What does the call start distribution look like for yesterday?
Show pickup rates by phone number this month
Which outbound number has the best pickup rate?
Show phone number trends for the last 2 weeks
What are the call duration stats this month? (avg, median, p95)
Show attempt cohort analysis — how does success rate change with retries?
Give me today's daily summary

Organization

Invite alice@company.com to my org
Invite bob@acme.com and carol@acme.com as admins
Invite these emails as members: dev1@company.com, dev2@company.com, dev3@company.com

Audiences

List all my audiences
Show me the members of audience "Sales Leads"
Search for +14155551234 in the "Sales Leads" audience
Add these contacts to the "Sales Leads" audience:
phoneNumber: +14155551234, firstName: John, company: Acme
phoneNumber: +14155559876, firstName: Jane, company: Beta Corp
Remove the member with phone +14155551234 from "Sales Leads"
Delete the "Old Test Audience" audience

Campaigns

Starting a campaign triggers real phone calls to every contact in the audience.

Show me all running campaigns
List completed campaigns
Get details for campaign "January Outreach"
Create a campaign called "Product Launch" using agent "Sales Bot" and audience "Sales Leads"
Create a campaign "Follow Up" with agent "Support Bot", audience "Customers",
max 2 retries, 30 minute retry delay
Start the "Product Launch" campaign
Pause "Product Launch" — we need to fix the prompt first
Resume the "Product Launch" campaign
Export the call logs for campaign "January Outreach"
Export campaign "January Outreach" results as CSV
Delete the "Test Campaign" campaign

Phone Numbers

List all phone numbers in my organization
Which agents are assigned to which phone numbers?

Billing & Payments

What's my current credit balance?
Show me my credit usage breakdown by feature
Show my transaction history for the last 7 days
Show only usage deductions from this month
What plan am I on?
What features are available on my plan and what are the credit rates?
Show me all available plans and their features
List my invoices
What payment methods do I have on file?
What are my billing alert settings?
Set up billing alerts at $10, $50, and $100 thresholds
Add alerts@mycompany.com to billing notifications
Is auto-reload enabled?
Is coupon code WELCOME2025 valid?
Redeem coupon WELCOME2025

Chaining Actions

You can combine multiple steps in a single prompt — the AI executes them in sequence:

Create a new agent called "Demo Bot",
set its prompt to "You are a friendly demo agent that introduces
our product in under 30 seconds",
set the voice to yuvika,
then call +14155551234 with it

This will:

  1. Create the agent
  2. Update its prompt
  3. Change the voice
  4. Make the call

All in one go.

Power User Tips

What model and voice is "Sales Bot" using?
What languages does "Support Bot" support?
Is voicemail detection enabled on "Outreach Agent"?
List all agents that have 0 calls and archive them
Show me all agents using the electron model
Show me failed calls this week, then debug the most recent one
Which agent had the most failed calls this month? Show me one of the failure transcripts.
Compare call volume this week vs last week