Can the assistant sign up or manage billing?

It can start the signup itself with the keyless signup tools and later fetch the hosted payment link, but a human always closes the loop: signing in, verifying the phone number, accepting terms, and paying happen in person, in the dashboard or on the hosted payment page. The assistant never touches payment credentials.

Integration

Your AI assistant can operate your phone system

Q: Can ChatGPT or Claude access my Canarics account?

Yes. A team manager creates a read-only API key in the Canarics dashboard and gives it to the assistant. Any AI tool that can make HTTP requests can then check minute usage, read call summaries with scores, and pull daily statistics. Claude connects in one line as a remote MCP server.

Q: Is there an MCP server for phone calls and call analytics?

Yes. Canarics is a remote MCP server (Streamable HTTP) listed in the official MCP registry as com.canarics/canarics. It gives an AI assistant voice agent management, call analytics, caller history, and consent-gated callbacks over a business phone line, plus keyless signup tools for companies without an account yet.

Q: Can my AI assistant make calls through Canarics?

It can request a callback and draft calling campaigns. On the outbound plan, with a key a manager has explicitly enabled, the assistant can ask the team's Canarics voice agent to call a known customer back, or prepare a whole outbound campaign as a disabled draft that a manager reviews and launches in the dashboard. Callbacks only work toward numbers with existing call history, campaign drafts never place calls by themselves, and all dialing runs through the same business-hours and billing safeguards as every other call. It is never a cold-calling API.

Q: Is it safe to give an AI assistant access to call data?

Keys are read-only by default, scoped to one team, revocable in one click, and stored only as cryptographic hashes. Responses go through strict field allowlists: no transcripts, no integration tokens, no other team's data. Everything is processed and hosted in the EU.

More and more companies have their own AI assistant: a Claude connected to company tools, a ChatGPT with actions, an internal agent watching operations. Canarics is built so that assistant can work with your phone line directly: read every call result, notice a missed customer the moment it happens, and ask your voice agent to call them back.

What your assistant can do

Capability	How
Check minute usage	Minutes used against the plan this billing cycle, and the cap where calls stop. No surprise invoices.
Read call results	AI-written summaries, quality scores, sentiment, and missed or recovered status for every call, human or AI-handled. Transcripts stay private by design.
Pull daily statistics	Per-day totals, missed and recovered counts, and score averages for up to 90 days, in your timezone.
Look up a caller	Cross-call history for one number: how many touches, the last outcome, recent conversation summaries.
React in real time	Canarics pushes signed webhook events to your systems: call.analyzed after every analyzed call, call.missed the moment an inbound customer call goes unanswered.
Request a callback	On the outbound plan, with a manager-enabled key, the assistant schedules your Canarics voice agent to call a known customer back, with a reason grounded in what the customer asked.
Draft a calling campaign	On the outbound plan, with a manager-enabled key, the assistant prepares an outbound campaign as a disabled draft: audience, briefing, daily limit, calling window. It appears in your dashboard marked "AI draft"; only a manager can launch it. Drafting never places a call.
Set up Canarics from scratch	No account needed: the assistant starts the signup itself (no key), sends you one claim link to verify your identity and phone, and receives its own API key once the trial is live.
Ring you a live demo	On a trial, ask your assistant to hear the AI in action: it triggers a demo call and your phone rings within seconds. Always your own verified number, never anyone else's, capped by the trial's call limit.
Fetch the payment link	On a trial, the assistant creates the paid-plan checkout with your company billing details and hands you the hosted payment link. You pay; activation is automatic. The assistant never touches payment credentials.

Onboarding, agent-first

Tell your assistant "set up AI call handling for us" and it can do almost all of it. It calls POST /api/v1/signup (or the keyless start_signup MCP tool) with your company name and website, and gets back a claim link. You open that link once: sign in, verify your phone by SMS, accept the terms. The trial provisions itself in seconds, with a demo voice agent already written from your website. Your assistant follows along on the status endpoint and picks up a read-only API key for the new team the moment the trial is live; from there it reads your calls, and when you are ready it fetches the upgrade payment link for you to pay.

The boundary is deliberate: identity, phone ownership, terms, and payment are always yours. Everything an assistant can safely do, it can do alone.

Connect your assistant

Don't set anything up yourself: tell your assistant to read canarics.com/llms.txt and connect to Canarics. It finds the endpoints, the tools, the signup flow, and the exact connector settings there on its own, and tells you if it needs anything from you.

The MCP server for phone calls and call analytics

Everything above is also a remote MCP server, listed in the official MCP registry as com.canarics/canarics and in the ChatGPT connector directory. It speaks Streamable HTTP, is stateless, and needs no installation: any MCP-capable client connects to the endpoint directly. With an API key it exposes the voice agent and call analytics tools (usage, agents, call summaries, conversations, daily statistics, caller lookup, callbacks, campaign drafting); with no key at all it exposes the signup tools, so an assistant can set a brand-new company up before any account exists.

In Claude Code, one line:

claude mcp add --transport http canarics https://canarics.com/mcp --header "Authorization: Bearer cnr_..."

Use Canarics in n8n, no custom node needed

n8n speaks MCP natively, so Canarics works in any n8n AI Agent workflow today:

Add the MCP Client Tool node to your AI Agent workflow.
Endpoint: https://canarics.com/mcp. Server Transport: HTTP Streamable.
Authentication: Bearer Auth, with a credential holding your cnr_ API key from the dashboard Team tab. With no credential at all, the agent still gets the keyless signup tools.
Tools to Include: All. Your agent now sees usage, calls, conversations, daily stats and caller lookups as tools.

For triggers, pair it with n8n's own Webhook node: paste the webhook URL into the Canarics dashboard Team tab, and every call.analyzed and call.missed event starts a workflow. Deliveries are HMAC-signed (X-Canarics-Signature), so the workflow can verify each event really came from Canarics. The same recipe works in Make, Zapier or any platform that can receive an HTTPS request; n8n is simply the one with native MCP support built in.

The loop nobody else closes

Put the pieces together and two AI systems handle a missed customer with no human in the critical path:

A customer calls, nobody picks up. Within seconds a call.missed event lands on your endpoint.
Your assistant looks up the caller's history: four previous conversations, last one about a price quote.
It requests a callback with that context: "Customer was waiting for the winter tire quote."
The Canarics voice agent calls the customer back, briefed, in the customer's language.
The analyzed result returns to your systems as a call.analyzed event, summary and score included.

Your assistant decides; our agent speaks. Each side does what it is good at.

Security is the design, not a setting

Giving an external AI access to call data deserves paranoia, so the defaults are strict. Keys are read-only unless a manager explicitly enables callbacks or campaign drafting, scoped to one team, revocable in one click, and stored only as cryptographic hashes. Every response passes a field allowlist: no transcripts, no integration tokens, never another team's data. Callback requests only work toward numbers that have already talked with your team, capped per day, inside calling hours, and the callback reason is treated as data, never as instructions to the voice agent. Campaigns drafted by an assistant are proposals: they stay disabled until a manager reviews and launches them in the dashboard, so there is no path from an assistant's context to phones ringing without a human decision. Everything runs and stays in the EU.

Common questions

Can ChatGPT or Claude access my Canarics account?

Yes. A team manager creates a read-only API key in the dashboard and gives it to the assistant. Any AI tool that can make HTTP requests can then check usage, read call summaries, and pull statistics. Claude connects in one line as a remote MCP server.

Is there an MCP server for phone calls and call analytics?

Yes, this is one. Canarics is a remote MCP server (Streamable HTTP) listed in the official MCP registry as com.canarics/canarics. It gives an AI assistant voice agent management, call analytics, caller history, and consent-gated callbacks over your business phone line, plus keyless signup tools for companies without an account yet.

Can my AI assistant make calls through Canarics?

It can request a callback to a customer your team already has history with, and it can draft whole outbound campaigns: audience, briefing, daily limit, ready to go. Drafts stay disabled until a manager reviews and launches them in the dashboard, so the assistant prepares and a human decides. Volume is capped, calling hours are enforced, and the same billing safeguards apply as to every other call. It is never a cold-calling API.

Does Canarics work with n8n, Zapier or Make?

n8n works today with nothing to install: the MCP Client Tool node connects to https://canarics.com/mcp over HTTP Streamable with Bearer auth, and an n8n Webhook node receives call.analyzed and call.missed events as workflow triggers. Zapier and Make can use the same webhooks plus the REST API through their generic HTTP modules.

Is it safe to give an AI assistant access to call data?

Keys are read-only by default, team-scoped, revocable, and stored only as hashes. Responses carry no transcripts and no credentials. If a key leaks, the worst case is disclosure of your own team's call summaries until you revoke it, and never an action that costs money.

Where do I start?

Two ways. Hands-on: start a free trial, open the Team tab, create an API key, and hand it to your assistant together with the docs link. Assistant-first: tell your AI assistant to set Canarics up for you, it starts the signup itself and sends you one link to claim. Assistants find their way alone: this site describes the API in llms.txt, the format AI tools read.

Your AI reads the docs itself.
Paste canarics.com/openapi.json into your assistant and ask it what it could do for your phone line.

Request a demo