MCP Server

Run Speka from your AI assistant.

Speka ships a Model Context Protocol server. Connect it to Claude, Cursor, ChatGPT, Hermes AI, OpenClaw AI, n8n, Zapier, Make or any MCP client and your assistant can manage your account in plain language — create agents, edit prompts & knowledge, choose voices, connect tools, query your agents' knowledge, and review conversations, leads, recordings & analytics.

Get your connection URL →Install guide

Transport: Streamable HTTP (JSON-RPC over POST) · Auth: bearer token

Key takeaways

The Speka MCP server lets Claude, Cursor, ChatGPT, Hermes AI and OpenClaw AI manage your Speka voice agents by chat.
Connect in 3 steps: generate a secure URL in Dashboard → MCP Server, paste it into your client, then just ask.
Auth is a bearer token — embedded in the URL, or sent as an Authorization: Bearer header.
Your assistant can view, create, edit & delete agents, list voices, connect tools, and read recordings, transcripts & feedback.
Transport is Streamable HTTP (JSON-RPC over POST); works with any MCP-compatible client.

What is the Speka MCP server?

The Speka MCP server is a hosted Model Context Protocol endpoint that lets AI assistants like Claude, Cursor and ChatGPT manage your Speka voice and text agents through natural-language commands. You generate a secure URL with an embedded bearer token, paste it into your client, and your assistant can then create agents, edit prompts, select voices, connect tools, and read call recordings, transcripts and feedback — all within the scopes you allow. It uses Streamable HTTP (JSON-RPC over POST) and works with any MCP-compatible client. The Model Context Protocol is the open standard introduced by Anthropic for connecting AI assistants to external tools and data.

Which AI assistants support MCP?

Speka's MCP server works with any client that supports a remote Streamable-HTTP MCP server. Confirmed clients include Claude Desktop, Claude.ai (web), Cursor, VS Code, ChatGPT / the OpenAI Responses API, Hermes AI, and OpenClaw AI. Any other compliant client connects with the same URL — verify it with the cURL test in the install guide below. Speka also has flat pricing with no per-minute fees.

Keep your token secret

Your bearer token is like a password: anyone who has it can act on your account within the scopes you grant. Embed it in the URL for personal use, or send it as an Authorization: Bearer header in shared environments — and regenerate it any time in Dashboard → MCP Server to instantly revoke the previous token.

Last updated: 5 June 2026

How it works

Three steps to a connected assistant.

Generate a connection

In the dashboard, open MCP Server and click Generate. You get a secure server URL with your token embedded.

Paste it into your client

Add it to Claude, Cursor, ChatGPT, Hermes AI, OpenClaw AI, n8n, Zapier or Make — any MCP client — using the snippet for your tool below.

Just ask

Your assistant can now create agents, edit prompts & knowledge, pick voices, connect tools, ask an agent against its knowledge, and read conversations, leads, recordings & analytics — within the scopes you allow.

Your connection

One secure URL.

Generate yours in Dashboard → MCP Server. It looks like this — the token is part of the URL, so most clients need nothing else:

http://127.0.0.1:4317/api/mcp/YOUR_TOKEN

Prefer a header? Use the endpoint http://127.0.0.1:4317/api/mcp with Authorization: Bearer YOUR_TOKEN. Keep your token secret — anyone with it can act on your account within its scopes. Regenerate any time to revoke the old one.

Install

Connect your client.

Replace YOUR_TOKEN with the token from your dashboard (or copy the pre-filled snippet straight from the MCP Server tab).

Claude Desktop

Easiest (no install): Settings → Connectors → Add custom connector → paste the Server URL. Only the classic config-file method below needs Node.js (it bridges via mcp-remote):

{
  "mcpServers": {
    "voice-studio": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://127.0.0.1:4317/api/mcp",
        "--header",
        "Authorization: Bearer YOUR_TOKEN"
      ]
    }
  }
}

Claude.ai (web)

Settings → Connectors → Add custom connector → paste the Server URL.

http://127.0.0.1:4317/api/mcp/YOUR_TOKEN

Cursor

Create or edit ~/.cursor/mcp.json, then reload Cursor.

{
  "mcpServers": {
    "voice-studio": {
      "url": "http://127.0.0.1:4317/api/mcp/YOUR_TOKEN"
    }
  }
}

VS Code

Add to .vscode/mcp.json in your workspace (or run “MCP: Add Server”).

{
  "servers": {
    "voice-studio": {
      "type": "http",
      "url": "http://127.0.0.1:4317/api/mcp/YOUR_TOKEN"
    }
  }
}

ChatGPT / OpenAI Responses API

Add as an MCP tool in the OpenAI Responses API (or a ChatGPT custom connector).

{
  "type": "mcp",
  "server_label": "voice-studio",
  "server_url": "http://127.0.0.1:4317/api/mcp/YOUR_TOKEN"
}

Hermes AI

Add to ~/.hermes/config.yaml (or run `hermes mcp add`) — connects directly over HTTPS, no install needed.

mcp_servers:
  speka:
    url: "http://127.0.0.1:4317/api/mcp"
    headers:
      Authorization: "Bearer YOUR_TOKEN"

OpenClaw AI

Add to mcporter.json — Streamable HTTP, connects directly over HTTPS.

{
  "mcpServers": {
    "speka": {
      "type": "streamable-http",
      "url": "http://127.0.0.1:4317/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

n8n

Add an MCP Client Tool node (sub-node of the AI Agent). Endpoint type = HTTP Streamable, Authentication = Bearer.

Endpoint:   http://127.0.0.1:4317/api/mcp
Transport:  HTTP Streamable
Auth:       Bearer Token
Token:      YOUR_TOKEN

Zapier

Apps → Add connection → MCP Client by Zapier. Transport = Streamable HTTP.

Server URL: http://127.0.0.1:4317/api/mcp
Transport:  Streamable HTTP
Bearer:     YOUR_TOKEN

Make.com

Add the MCP module and connect a remote MCP server (Streamable HTTP).

URL:    http://127.0.0.1:4317/api/mcp
Auth:   Bearer YOUR_TOKEN

Any other client / cURL test

Every MCP client that supports a remote Streamable-HTTP server uses the same URL. Verify it from a terminal:

curl -X POST http://127.0.0.1:4317/api/mcp \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Capabilities

The tools your assistant gets.

Each capability maps to a scope you can toggle on or off per connection.

View agents

agents:read

list_agents — List your agents (name, voice, prompt preview).
get_agent — Get one agent in full.
get_embed_snippet — Get the one-line website embed code.

Create & edit agents

agents:write

create_agent — Create a new voice agent.
update_agent — Edit name, prompt, first/welcome message, voice.
set_agent_voice — Set the agent’s voice.

Delete agents

agents:delete

delete_agent — Archive (delete) an agent.

View knowledge

knowledge:read

get_knowledge — Read an agent's knowledge base.

Edit knowledge

knowledge:write

set_knowledge — Replace the knowledge base.
add_knowledge — Append text to the knowledge base.

Ask agents

knowledge:query

ask_agent — Ask an agent a question against its knowledge.

List voices

voices:read

list_voices — List standard & premium voices.

Connect tools

tools:write

connect_tool — Attach a webhook tool the agent can call.

Read conversations

conversations:read

list_conversations — List text-chat conversations.
get_conversation — Full transcript of one conversation.
search_conversations — Keyword search across transcripts.

Read leads

leads:read

list_leads — List captured name/email leads.

Read recordings

recordings:read

list_recordings — List call recordings.

Read transcripts

transcriptions:read

list_transcriptions — List recordings with transcripts.
get_recording_transcript — Full transcript of one recording.

Read feedback

feedback:read

list_feedback — List call ratings & comments.

Analytics & account

analytics:read

get_analytics — Usage over a date range.
account_status — Plan + remaining quota.

In practice

Just talk to your assistant.

“List my voice agents and tell me which voice each one uses.”

“Create a new agent called “Sales Bot” with a friendly prompt and the Kore voice.”

“Ask my Support agent: “what is your refund window?” using its knowledge base.”

“Show me the leads captured this week and the last 5 chat conversations.”

Ready to connect?

Generate your MCP connection in the dashboard, paste it into your client, and start managing Speka by chat.

Open MCP Server →

Frequently asked questions

What is the Speka MCP server?

A hosted Model Context Protocol endpoint that lets AI assistants such as Claude, Cursor and ChatGPT manage your Speka voice and text agents in plain language. It runs over Streamable HTTP and uses a bearer token for authentication.

How do I connect Claude to the Speka MCP server?

In Claude Desktop, go to Settings → Connectors → Add custom connector and paste your Speka Server URL — no install needed. The mcp-remote config-file method is only required if you use the classic Node.js setup.

Can I use ChatGPT to manage my Speka agents?

Yes. Add Speka as an MCP tool in the OpenAI Responses API, or as a ChatGPT custom connector, using your server_url with the embedded token. Your assistant can then create and edit agents directly.

How do I generate a secure MCP connection URL?

Open Dashboard → MCP Server and click Generate. You get a URL with your bearer token embedded. Keep it secret — anyone with it can act on your account within its scopes — and regenerate any time to revoke the old one.

Which clients support the Speka MCP server?

Any client that supports a remote Streamable-HTTP MCP server: Claude Desktop, Claude.ai (web), Cursor, VS Code, ChatGPT / OpenAI Responses API, Hermes AI, OpenClaw AI, and any other compliant client (verify with the cURL test).

What can my assistant actually do once connected?

Within the scopes you allow, it can view, create, edit and delete agents, list available voices, connect tools, and read call recordings, transcripts and feedback — all in plain language.