MCP server

Voxa runs a Model Context Protocol server. Once an assistant is connected, you can ask it things like:

• “Create a receptionist agent for my clinic that speaks Hindi and books appointments.”
• “Lower the temperature of the booking agent and show me what changed.”
• “Why did yesterday’s calls to the support agent fail?”
• “Call +91 98123 45678 with the reminder agent and tell me how it went.”

URL	https://voxa.abhinavyadav.in/mcp
Transport	Streamable HTTP
Sign-in	OAuth 2.1 (recommended), or an API key
Tools	18: agents, versions, calls, timelines, billing, limits, voices, models, tool testing (full list)

How sign-in works

Most assistants connect with OAuth: you add the URL, the assistant opens a Voxa page in your browser, you sign in and press Allow. No key is copied anywhere.

1. The assistant calls /mcp and gets a 401 that says where to sign in.
2. It registers itself with Voxa and opens the consent page (/oauth/authorize).
3. You sign in, check the workspace and what the app will be able to do, and choose Allow.
4. The assistant receives an access token (valid for 1 hour) and a refresh token (valid for 30 days), and renews them on its own.

The connection acts as you, in the workspace you approved, with your current role, but never more than an API key could do. It can’t change billing, members or API keys. If you leave the workspace or your role changes, the connection follows at once.

Approve only apps you started connecting yourself. The consent page shows the app’s name and the address it will return to.

Disconnecting

Open Settings → Connected apps in the console and choose Disconnect. The app loses access immediately. Connect it again from the app if you want it back.

With an API key instead

Clients that can’t do OAuth, scripts, and CI jobs can send a workspace API key instead: Authorization: Bearer vx_... (or X-API-Key: vx_...). A key acts as a developer of its workspace. Create one under API keys in the console, and keep it private: anyone with the key can act in your workspace. See Authentication.

What the tools can do

Every tool goes through the same checks as the REST API: permissions, plan limits, credits and rate limits. Each tool run is written to your audit log as mcp.<tool>, next to the change itself, with the actor shown as “your name via app name”.

Only tools that just read are marked read-only. Many assistants run read-only tools without asking, so anything that changes your workspace, costs credits, or sends a request to the internet will always ask you first.

Next steps

• Quickstart: connect Claude Code, Claude Desktop, Cursor, VS Code, Windsurf or another client.
• Tools: every tool, its arguments and what it returns.
• Prompt cheatsheet: prompts to copy, grouped by task.