MCP server: run Call2Me from ChatGPT or Claude

The MCP server exposes your Call2Me account to an AI assistant as a set of tools. Once connected, you ask for what you want — "read me the transcript of my last call", "buy a US number for this agent" — and the assistant calls the API on your behalf.

Server URL:

Copyhttps://mcp.call2me.app/mcp

Connecting

1. In your MCP client, add a connector and paste the server URL above.
2. Your browser opens the Call2Me connect page. Paste your API key once (create one at dash.call2me.app → API Keys) and approve the scopes you want.
3. Ask for something. Start with "list my agents" — it is read-only, so nothing can go wrong while you look around.

The key is pasted into our page, not into your AI client: it is stored encrypted on our side and never reaches the model. The client only receives a short-lived OAuth token bound to your account. No config file to edit.

Claude (web and desktop)

Customize → Connectors → + Add → Add custom connector, then paste the server URL. Claude opens the Call2Me connect page, you paste your key and approve the scopes, and the tools appear in the conversation. On Team and Enterprise plans an owner adds the connector first. The desktop app uses the same connector — no claude_desktop_config.json needed.

Call2Me is published in the official MCP registry as app.call2me/mcp, so clients that read the registry can find it by name instead of a pasted URL.

Claude Code

Copyclaude mcp add --transport http call2me https://mcp.call2me.app/mcp

Then run /mcp inside Claude Code, pick call2me and choose Authenticate to open the connect page.

ChatGPT

Settings → Apps & Connectors → enable Developer mode, then Create and paste the server URL. Menu names vary by plan; on some plans developer-mode connectors are read-only. Nothing needs approving on our side — the connection is between your ChatGPT account and our server.

Cursor

~/.cursor/mcp.json (or .cursor/mcp.json in a project):

Copy{ "mcpServers": { "call2me": { "url": "https://mcp.call2me.app/mcp" } } }

VS Code

.vscode/mcp.json:

Copy{ "servers": { "call2me": { "type": "http", "url": "https://mcp.call2me.app/mcp" } } }

Windsurf

~/.codeium/windsurf/mcp_config.json:

Copy{ "mcpServers": { "call2me": { "serverUrl": "https://mcp.call2me.app/mcp" } } }

MCP Inspector

Copynpx @modelcontextprotocol/inspector --server-url https://mcp.call2me.app/mcp --transport http

Start the OAuth flow from the Inspector's authentication panel.

Other clients

Any client that speaks Streamable HTTP works the same way: give it the URL and complete the browser connect step. Clients may register dynamically (/register) or with a Client ID Metadata Document. We negotiate protocol versions 2025-03-26, 2025-06-18, 2025-11-25 and 2026-07-28.

Scopes

Granting only call2me:read is a real restriction, not a hint: the server rejects every write with a 403 regardless of what the assistant asks for. Useful for reporting, and for trying it out before you let it touch anything.

The tools

The server currently lists 72 tools: 34 read-only, 18 write and 20 destructive. Every tool declares what it does, so a compliant client can decide when to check with you first. The three annotations are set explicitly on all of them — the MCP spec defaults destructiveHint and openWorldHint to true, so leaving them unset would mark harmless read tools as dangerous and bury the genuinely destructive ones in the noise.

Read-only

call2me_profile · call2me_get_wallet · call2me_get_pricing · call2me_list_agents · call2me_get_agent · call2me_list_calls · call2me_get_call · call2me_get_call_transcript · call2me_list_phone_numbers · call2me_search_numbers · call2me_list_allowed_countries · call2me_list_schedules · call2me_list_interpreters · call2me_get_interpreter · call2me_list_interpreter_calls · call2me_get_interpreter_web_status · call2me_list_knowledge_bases · call2me_get_knowledge_base · call2me_query_knowledge_base · call2me_list_widgets · call2me_get_widget · call2me_list_chats · call2me_get_chat · call2me_get_schedule · call2me_get_agent_stats · call2me_get_call_recording · call2me_list_voices · call2me_list_sip_trunks · call2me_list_sms · call2me_list_transactions · call2me_get_usage_analytics · call2me_get_auto_charge · call2me_get_webhook · call2me_list_notification_targets

Write

call2me_create_agent · call2me_update_agent · call2me_create_schedule · call2me_bind_number · call2me_checkout_number · call2me_create_voice_session · call2me_create_interpreter · call2me_update_interpreter · call2me_enable_interpreter_web · call2me_create_interpreter_passcode · call2me_create_knowledge_base · call2me_add_knowledge_base_sources · call2me_create_widget · call2me_update_widget · call2me_start_chat · call2me_update_schedule · call2me_duplicate_agent · call2me_create_topup_link (returns a payment link, never pays)

Destructive

These spend money, cannot be undone, or send your calls and data to a new destination. A compliant client asks you first.

call2me_make_outbound_call · call2me_end_call · call2me_purchase_number · call2me_release_number · call2me_unbind_number · call2me_delete_agent · call2me_cancel_schedule · call2me_start_interpreter_call · call2me_delete_interpreter · call2me_end_interpreter_web · call2me_send_sms · call2me_send_chat_message (billed per message) · call2me_delete_knowledge_base · call2me_delete_widget · call2me_delete_sip_trunk · call2me_delete_notification_target · call2me_create_sip_trunk · call2me_register_phone_number (requires your own SIP trunk) · call2me_set_webhook · call2me_create_notification_target

Wallet auto top-up settings can be read (call2me_get_auto_charge) but are changed only in the dashboard — an assistant cannot set up automatic card charges.

Interpreter support

The interpreter is fully covered — eleven of the tools above are interpreter tools. From an assistant you can list and inspect interpreters, create a language pair, enable the browser room, mint a single-use passcode, start a phone session and end a live browser session. See Live interpreter for what the product does and the API reference for the raw endpoints.

Protocol details

All four protocol revisions are served because the transport changed between them; your client negotiates whichever it speaks via the MCP-Protocol-Version header.

What it does not do

• It cannot read other tenants' data. Tools run with your account's own permissions, same as your API key.
• It cannot mint more tokens. The access token reaches the MCP tools and nothing else.
• It does not bypass your balance. Calls and numbers draw on the same wallet as the dashboard; an insufficient balance fails the tool call rather than going into debt.