MODEL CONTEXT PROTOCOL

Give your AI a phone line.

The hosted Sautikit MCP server connects Claude Code, Cursor, VS Code, and Windsurf straight to your workspace. Your AI reads calls, wires numbers, builds voice agents, and launches campaigns — and you approve anything that spends money.

Connect in 30 secondsRead the setup guide

One endpoint. Your existing API key.

No new accounts, no SDK, no local process. Point your client at https://mcp.sautikit.com/mcp with a Sautikit API key and 42 tools appear.

claude mcp add --transport http sautikit https://mcp.sautikit.com/mcp \
  --header "Authorization: Bearer $SAUTIKIT_API_KEY"

42 tools, grouped the way you work.

Every tool wraps the public REST API — the same auth, scopes, rate limits, and billing as your own code. Read tools are marked read-only; destructive and money-spending tools carry annotations so your client asks you first.

Agents

12 tools

Draft, publish, and version AI voice agents. Attach HTTP tools. Place one-off agent calls.

list_agents · get_agent · create_agent · update_agent · delete_agent · publish_agent · list_agent_revisions · list_agent_tools · upsert_agent_tool · delete_agent_tool · send_agent_call · get_agent_limits

Broadcasts

8 tools

Build outbound campaigns, load contacts, launch with your confirmation, read the report.

list_broadcasts · get_broadcast · get_broadcast_report · create_broadcast · add_broadcast_contacts · start_broadcast · pause_broadcast · resume_broadcast

Calls

6 tools

List and inspect calls with full event timelines, pull recordings, place and hang up calls.

list_calls · get_call · get_call_stats · get_call_recording · place_call · hangup_call

Numbers

6 tools

Search available numbers, claim and release them, and rewire routing on the fly.

list_numbers · get_number · search_available_numbers · claim_number · release_number · update_number_routing

Webhooks

6 tools

Create and test event subscriptions, then debug deliveries without leaving your editor.

list_webhooks · list_webhook_deliveries · create_webhook · update_webhook · test_webhook · delete_webhook

Wallet

3 tools

Balance, statements, and AI spend — so your agent always knows what a plan will cost.

get_wallet_balance · list_wallet_statements · get_ai_spend

Authoring

1 tools

Validate voice-action documents against the real schema before anything touches a live number.

validate_voice_actions

Wire up an IVR in one conversation.

Tell your AI what the menu should do. It authors the voice-action JSON, proves it against the real schema with validate_voice_actions — hallucinated fields die here, not on a live call — then points your number at it with update_number_routing.

ivr.json
{
  "actions": [
    {
      "getDigits": {
        "numDigits": 1,
        "timeout": 10,
        "nested": [
          {
            "say": {
              "text": "Welcome to Wema Dental. Press 1 to book an appointment, press 2 for opening hours."
            }
          }
        ]
      }
    },
    { "say": { "text": "Sorry, we didn't catch that. Goodbye." } },
    { "hangup": {} }
  ]
}

Callers press a digit; the digits are posted back to your webhook, which decides what happens next. Validation is local and free — no call, no charge, no surprises.

From a one-line brief to a campaign report.

“Call our 200 July appointments and remind them” is now a complete instruction. The whole pipeline runs over MCP, end to end:

  1. create_agentYour AI drafts the agent: prompt, voice, model — from a one-line brief.
  2. publish_agentIt freezes the draft as an immutable revision, so the campaign can't drift mid-flight.
  3. create_broadcastIt sets up the campaign: template with {{variables}}, schedule window, retry policy.
  4. add_broadcast_contactsIt loads your contact list and reports every rejected row with the exact reason.
  5. start_broadcastDialing starts only after YOU confirm — this tool is flagged as money-spending.
  6. get_broadcast_reportWhen the campaign completes, it pulls the funnel: reached, no-answer, busy, cost.

Guardrails are the product.

Tools are gated by your API key’s scopes — mint a narrow key per agent and it only sees the tools it can use. Spending and destructive tools require an explicit confirmation in your client. Secrets never round-trip: tool headers are write-only, and key minting, payments, and secret rotation are not exposed as tools at all.

Put a voice stack in your AI’s hands.

Mint an API key in the dashboard, paste one command, and ask your AI to call you.

Get your API keySetup guide