VideoCallTranslator MCP connector reference

What does the connector do?

It lets an AI assistant prepare translated video calls on a host's VideoCallTranslator account: create a call room and return its two links, list the host's rooms, report the minutes left, and list the call languages.

The call itself happens on videocalltranslator.com in the browser, where the host opens the host link and the other person opens the guest link. The connector cannot start, join or listen to a call, sends no invitations, and takes no payments. Minutes are bought on the pricing page.

For how a person connects it, see how to set up a translated video call from Claude.

How does an assistant connect?

• Server: https://videocalltranslator.com/mcp, Streamable HTTP, stateless. It serves the 2026-07-28 MCP revision and the 2025 revisions.
• Directory: listed in Claude's connector directory as a Community connector.
• Authorization: OAuth 2.1, authorization code with PKCE (S256). Discovery documents are at /.well-known/oauth-protected-resource/mcp and /.well-known/oauth-authorization-server.
• Client registration: a Client ID Metadata Document, or dynamic client registration at /oauth/register. No client secret has to be entered by the person connecting.
• Sign-in: the person signs in on videocalltranslator.com, with Google or with email and password, then answers our approval page, which names the assistant, and its domain when the assistant proves one.
• Tokens: access tokens last an hour and are renewed with a refresh token. A connection unused for 30 days expires.

Which permissions can the person grant?

All three are offered unless the assistant asks for fewer. On the approval page, room creation is a ticked box: unticking it connects the assistant read-only, with rooms.read and balance.read, and the account page then marks that connection Read-only. An assistant can also ask for read-only access itself, with scope=rooms.read balance.read.

What are the tools?

No tool is a sensitive write. None pays, buys, sends a message, or deletes anything. create_call_room adds a room, which is free; translation minutes are used only while two people are connected on a call.

Without host_language, a room takes the language set on the host's account. Without guest_language, the guest chooses theirs on the screen before joining. Every result has a text form and a structured form. No result contains an email address, a guest's name, a call record, audio, a transcript or payment details.

Each tool carries a title and the MCP annotations readOnlyHint, destructiveHint (false for all four) and openWorldHint (false).

What errors can a tool return?

• No token, or an expired one: HTTP 401, with the discovery document's address in WWW-Authenticate.
• A tool the person did not allow: HTTP 403 insufficient_scope, naming the scope it needs, so the assistant can ask for it.
• Invalid input: a tool error describing the field, for an unknown language code, a room name over 60 characters, or an unknown argument.
• Too many rooms in an hour: a tool error saying the limit was reached and that existing rooms still work (VCT-6020).
• Too many calls: a tool error giving the seconds to wait.

What are the limits?

• 60 tool calls a minute per account, across every connected assistant.
• The website's rooms-per-hour limit applies to create_call_room as well.
• Room names of at most 60 characters through the connector.
• Languages from the 41 call languages only.
• list_my_rooms returns at most 50 rooms.

How is a connection ended, and what is kept?

The host presses Disconnect under Account → Settings → Connected assistants, which ends the assistant's access at once. A connection unused for 30 days expires, and deleting the account ends every connection.

About a connection we keep which assistant it is and when it was connected, when it was last used (to the hour), and a record of each room it created. Its tokens are stored only in hashed form. What the assistant itself receives is kept by its own company under its own policy. The details are in our privacy policy.