Connect the Speechify MCP server

Speechify hosts a Model Context Protocol (MCP) server at https://mcp.speechify.ai/mcp. Connect it to your AI coding tool to ask grounded questions about the Speechify API, SDKs, code samples, and docs — with citations — without leaving your editor.

What you get

Set up with the Speechify CLI

If you use the Speechify CLI, it detects your installed clients and writes the config for you:

speechify mcp install --client claude-code   # or: cursor | vscode | windsurf | claude-desktop
speechify mcp install --all                  # every detected client
speechify mcp install --print                # print the config, write nothing

To configure a client by hand, use the sections below.

Claude Code

claude mcp add --transport http ask-speechify https://mcp.speechify.ai/mcp

Cursor

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

{
  "mcpServers": {
    "ask-speechify": {
      "type": "http",
      "url": "https://mcp.speechify.ai/mcp"
    }
  }
}

VS Code

Add the server to your VS Code mcp.json (used by Copilot):

{
  "servers": {
    "ask-speechify": {
      "type": "http",
      "url": "https://mcp.speechify.ai/mcp"
    }
  }
}

Windsurf

Add the server to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "ask-speechify": {
      "type": "http",
      "url": "https://mcp.speechify.ai/mcp"
    }
  }
}

Claude Desktop

Add the server to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "ask-speechify": {
      "type": "http",
      "url": "https://mcp.speechify.ai/mcp"
    }
  }
}

Config file locations:

{
  "mcpServers": {
    "speechify": {
      "command": "speechify",
      "args": ["mcp"]
    }
  }
}

Plain HTTP, no MCP client

If your agent can make an HTTP request but is not wired to speak MCP, query the retrieval engine directly. A GET to https://mcp.speechify.ai/v1/search returns the same ranked source passages the search tool returns, as JSON. There is no synthesis — you get the grounding, and your own model does the reasoning. No key is needed, and the endpoint is rate-limited to 30 requests per minute per IP.

import requests
response = requests.get(
    "https://mcp.speechify.ai/v1/search",
    params={"q": "which output_format should I use for telephony?", "limit": 5},
    timeout=30,
)
response.raise_for_status()
for hit in response.json()["hits"]:
    print(f"{hit['score']:.2f}  {hit['title']}  {hit['url'] or '(reference)'}")
    print(hit["body"][:200], "\n")

The response echoes the query and returns the passages best first:

{
  "query": "which output_format should I use for telephony?",
  "hits": [
    {
      "id": "…",
      "title": "Create Speech (Build › API Reference › Text to Speech)",
      "url": "https://docs.speechify.ai/build/api-reference/v1/audio/speech",
      "source": "db",
      "score": 0.72,
      "body": "…"
    }
  ]
}

Each hit carries a url you can cite when the passage comes from a public page. Reference passages drawn from the API specification have no page of their own, so their url is null — quote the body, and cite the API reference instead. Bodies are truncated at 1,200 characters; fetch the url (append .md for clean Markdown) when you need the whole page.

Verify the connection

Restart your client, then ask it something only the docs would know, for example:

How do I stream text-to-speech audio with the Speechify API?

The client should call the ask tool and answer with citations to the Speechify docs.