MCP

MCP (Model Context Protocol) lets an AI assistant work inside Speaknode on your behalf: read your calls, change an agent, build an analytics report. You connect it once, and from then on you ask the assistant in plain language instead of clicking through the interface.

What the assistant can reach

Area	Examples
Agents	Read the setup, change the prompt, switch the model or voice
Calls and sessions	Find a call, read the transcript, look at how it ended
Analytics	Run an analysis, read the scores, assemble a report
Tools	List the agent's tools, create one, run a test call against it
Tests	Start a test run, read the results
Reference data	Models, languages, staff and clients

The assistant sees exactly what you see. It works under your account, inside your space, with your permissions — it cannot reach a space you have no access to, and it cannot do anything the interface would not let you do.

What you need

The server address: https://test.api.speaknode.com/mcp
Your Speaknode account

There is nothing else to copy. No API key, no client ID, no secret — the address is the whole configuration.

How the connection works

You paste the address; the rest happens by itself:

The client asks the server who is allowed in and gets pointed at Speaknode.
A browser window opens with the usual login page.
You sign in and confirm access.
The client receives a token and starts working.

The token belongs to you personally. Your colleague connecting the same address logs in as themselves and sees their own data.

Connecting

Claude (web and desktop)

Open Settings → Connectors
Click Add custom connector
Paste https://test.api.speaknode.com/mcp as the server address
Confirm — a login window opens
Sign in with your Speaknode account

The connector appears in the list and stays available in every new chat.

Claude Code

claude mcp add --transport http speaknode https://test.api.speaknode.com/mcp

Then run /mcp inside Claude Code and pick Authenticate. A browser opens, you sign in, and the tools become available in the session.

Cursor

Add the server to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "speaknode": {
      "url": "https://test.api.speaknode.com/mcp"
    }
  }
}

Restart Cursor, open Settings → MCP, and click the sign-in prompt next to the server.

ChatGPT

Open Settings → Connectors
Click Add and choose a custom connector
Paste the server address and confirm
Sign in with your Speaknode account

Custom connectors are only available on paid ChatGPT plans. On a free plan the section will not appear.

Codex

Add the server to ~/.codex/config.toml:

[mcp_servers.speaknode]
url = "https://test.api.speaknode.com/mcp"

Restart Codex and confirm the sign-in when it asks.

Access and safety

The assistant acts as you. Every call it makes is recorded under your account, so anything it changes looks — correctly — like your change.

It can change things, not only read them. Creating an agent, editing a prompt, starting a test run cost money the same way they do from the interface. Read what the assistant proposes before you confirm it.

Revoking access — remove the connector in the client. Signing out of Speaknode in your browser does not disconnect it: the client holds its own token.

If something goes wrong

What you see	What it means	What to do
Login window does not open	The client did not reach the server	Check the address ends in `/mcp` and opens in a browser
`Invalid parameter: redirect_uri`	This client's callback address is not allowed yet	Send the full error text to support — one line has to be added on our side
Signed in, but no tools	The client connected before authentication finished	Reconnect the server in the client
Tools present, calls fail	Your account has no access to that space	Check in the interface that you can see the same data

MCP

On this page