Run your phone line
from your assistant.
Connect Call2Me to ChatGPT, Claude or any MCP client. Then just ask — place a call, read the transcript, set up an interpreter, buy a number. No API calls, no dashboard round-trip.
Things people actually ask it
What it can do — and what it asks first
Every tool declares whether it reads, writes or does something that cannot be undone. Your client uses that to decide when to check with you.
Read
Safe to run any time
Profile, agents, calls, transcripts, recordings, phone numbers, schedules, knowledge bases, widgets, chats, voices, SMS, wallet, transactions, analytics, webhook and notification settings, interpreters, pricing.
- “Read me the transcript of my last call”
- “Which of my numbers are active and what do they cost?”
- “How much credit is left?”
Write
Creates or changes something
Create and update agents, knowledge bases, widgets, and schedules; start chats, set up interpreters, register SIP trunks and numbers, set webhooks and notifications, change auto-charge, create a top-up link.
- “Create a Turkish receptionist agent for my clinic”
- “Schedule a follow-up call for tomorrow at 10:00”
- “Set up a Turkish↔German interpreter”
Sensitive
Spends money or cannot be undone
Place outbound calls, send SMS and paid chat messages, end a live call, purchase or release a number, delete agents, interpreters, knowledge bases, widgets or SIP trunks, cancel a schedule.
- “Call +90 532 … and ask about the invoice”
- “Buy a US number for this agent”
- “End the call that is running now”
Connecting it doesn’t hand over the keys
The assistant gets to use your account. It never gets to keep it.
Your API key never reaches the model
You sign in once with OAuth 2.1 (PKCE, S256). The key is encrypted at rest and used server-side only — the assistant sees results, never credentials.
Every tool declares what it does
Read-only, write and destructive tools are labelled separately, so your client can ask before anything spends money or deletes something.
Read-only access if you want it
Grant call2me:read and the connection can look but not touch — useful for reporting and for trying it out.
Revoke at any time
Disconnecting in your assistant ends the session. Refresh tokens rotate on every use and a reused token revokes the whole family.
Connect in three steps
Add the server in your assistant
Claude: Customize → Connectors → Add custom connector. Claude Code: claude mcp add --transport http call2me <url>. ChatGPT: Settings → Apps & Connectors → Developer mode → Create. Cursor, VS Code, Windsurf: add the URL to the MCP config. Then paste the server URL.
https://mcp.call2me.app/mcpPaste your key once and approve
Your browser opens the Call2Me connect page. Paste your API key (dash.call2me.app → API Keys) — it stays encrypted with us and never reaches the model — then approve read only, or read and write.
Ask for something
That’s it. Try “list my agents” first; it’s read-only, so nothing can go wrong while you look around.
The whole product, not just the reports
Calls
Place outbound calls, end a live one, read transcripts, schedule follow-ups.
Interpreting
Create interpreters, enable the browser room, issue a one-time passcode, start an interpreted call.
Numbers & agents
Search available numbers, check the price, buy, bind to an agent, release when you’re done.
Questions
What is MCP?
The Model Context Protocol is an open standard that lets an AI assistant use external tools. Connecting the Call2Me MCP server gives your assistant 72 tools for voice calls, interpreters and phone numbers — so you ask in plain language instead of calling the API yourself.
Which assistants can connect?
Any MCP client. We support both the 2025-11-25 and 2026-07-28 protocol revisions over Streamable HTTP, which covers ChatGPT, Claude and the common developer clients.
Does the assistant see my API key?
No. You authorise once through OAuth 2.1 with PKCE and the key is stored encrypted on our side. The model receives only the results of a tool call.
Can it spend my money without asking?
Placing calls and buying numbers are marked as sensitive tools, so a compliant client asks you first. You can also connect with read-only scope, which cannot write anything at all.
What does it cost?
The MCP server itself is free. You pay the normal per-minute rate for calls and interpreting, and the normal monthly price for any number you buy — the same prices as the dashboard.
Does it work with the interpreter?
Yes. You can list and create interpreters, enable browser interpreting, create a one-time passcode and start an interpreted call — all from the assistant.
Your voice AI, one sentence away
Free to connect. You only pay for the calls you make.