MCP call tools

The tools your AI uses for calls. The same server has every Kata Send tool too (company memory, lead lists, campaign status): call tools/list to see them all. import_leads and set_campaign_status also work on a call campaign.

create_call_campaign

Kata Call: create a phone campaign (status draft). An AI voice agent calls the leads, one engine session per call. Same leads, lists, brief.md and hours as an email campaign; its angles are call variants (opener + script + engine). Give opener and script to create the first variant (or use set_call_variant). Then: add_numbers, import_leads (csv with a phone column, campaign_id) or connect_list, and set_campaign_status active with the user's approval. Calls go out only from active numbers, in each lead's local calling hours.

update_call_campaign

Change a call campaign's settings: default engine and voice, agent name, max attempts, retry delays, max call length, calling hours, weekdays, country, autopilot.

add_numbers

Add phone numbers that place the calls (E.164, e.g. +33612345678). With real calls on (the workspace's switch on the Numbers page) a number must already be in the Twilio account (Kata never buys numbers); with real calls off (default) numbers are simulated and nothing dials. Default cap 50 calls a day per number.

list_numbers

Phone numbers with status (active, paused, retired), daily cap, calls today, live calls, dials and answer rate. Kata pauses a number by itself when its answer rate collapses (under 5 % on its last 30+ calls) and halves its cap under 15 %.

No parameters.

set_number

Pause, resume or retire a phone number, or change its daily cap or label.

set_call_variant

Create a call variant (campaign_id + opener and/or script) or change one (id). A variant = opener (said at pick-up) + script (goal, points, objections, the ask) + engine + voice. Thompson sampling picks one per call; success = interested, callback or a conversion. status retired removes it from the test. Scripts must only use facts from brief.md and company.md: speech models invent offers otherwise.

set_call_engine

Set the voice engine (grok or openai) of a call campaign (its default and all its live variants, unless only_default) or of one variant. To make the engines compete, keep variants on both engines: evolve also adds the best script on a missing engine.

import_call_leads

Import leads with phones into a call campaign from a CSV (phone column required: phone, phone_number, mobile or tel; email optional; first_name, company and any other column become facts the caller knows). National numbers use the row's country column, else the campaign's country. Leads on the do-not-call list are not enrolled.

place_test_call

Call a given number (the user's own) with a campaign's best variant or a given variant, to hear the agent. Not a lead, not in the stats. With real calls off nothing dials: the call is simulated (outcome forces the simulated result). With real calls on it REALLY CALLS: ask the user first.

list_calls

Calls, newest first, with filters: campaign_id, status (queued, ringing, in-progress, completed, no-answer, busy, failed, canceled, voicemail), outcome (interested, callback, not_interested, do_not_call, voicemail, wrong_person, no_conversation, or none), variant_id, number, contact_id, engine, since (ISO date), q (text), test (1 = test calls only, all = both). Each with duration, outcome and one-line summary. Page with before_id.

get_call

One call: status, timings, duration, outcome, summary, the transcript (turns of agent and lead), events (status changes, machine detection, engine, tool calls), the lead, and recording_url (an app URL, signed-in) when there is a recording.

call_stats

Kata Call performance, for one campaign or all: dials (today, total), answered and answer rate, voicemails, conversations, positive outcomes (interested + callback) and success rate, conversions and value, minutes, estimated cost, outcomes breakdown, per engine, per number, per variant (conversations, successes, rate, p_best = probability of being the best), and the last 90 days per day.

simulate_calls

Simulated calls (never dials, whatever the real-calls switch says): n waiting leads of a call campaign (or one contact) are called now by the simulator, ignoring hours and caps, through the real status flow and outcome tagging. outcome forces the result. For demos and to check a campaign end to end.

run_call_scheduler

Runs the call scheduler now, as the cron does every minute: active call campaigns, due leads inside their calling hours, one call per free active number. With real calls on this REALLY CALLS: ask the user first. dry_run shows what would be dialed without dialing.