MCP Servers

Lokutor runs two MCP servers. Both are hosted (Streamable HTTP), so adding one to Claude Code, Cursor, VS Code or any other MCP client is a single line. There is nothing to clone or install.

Server	URL	What it gives an AI assistant
Lokutor API	`https://api.lokutor.com/mcp`	Tools that call the API: voices, languages, models, request limits, status, and speech synthesis
Lokutor Docs	`https://docs.lokutor.com/mcp`	Search and read these docs, so answers about Lokutor come from the current documentation

Lokutor API

Tools

Tool	Does	API key
`list_voices`	`GET /voices`: every voice, with id, name, gender and languages	not needed
`list_languages`	`GET /languages`: the supported language codes	not needed
`list_models`	`GET /models`: loaded synthesis models and the default	not needed
`get_server_config`	`GET /config`: the limits a synthesis request is validated against	not needed
`get_server_status`	`GET /status`: readiness, version, uptime	not needed
`synthesize_speech`	`POST /tts/synthesize`: returns the audio (WAV, 44.1 kHz, mono) in the tool result	required; billed as text-to-speech, $4.10 per million characters, from your balance

synthesize_speech takes text (up to 1,000 characters), voice, language, speed (0.5–2.0) and steps (1–8, accepted but no effect: the current voice model runs a fixed 4). For longer text, call POST /tts/synthesize directly. Tools run exactly like the REST endpoints behind them, with the same authentication, balance checks and errors. A rejected key or a balance too low to start (billing.insufficient_balance) comes back as a tool error the assistant can read.

The Call Center API (buying numbers, placing outbound calls, bulk-dialing) is deliberately not exposed. Those endpoints spend real money and place real calls to real people, with consent_confirmed as the only built-in gate. The caller is the confirmation step, and an assistant calling a tool on its own is not that. Real-time voice agents run over WebSocket, which MCP tools do not model, so they are not tools either.

Add it

Create an API key in the dashboard and send it as the Authorization header. Without a key, the five discovery tools still work.

Claude Code
Cursor
VS Code
Claude (web and desktop)

claude mcp add --transport http lokutor https://api.lokutor.com/mcp \
  --header "Authorization: Bearer $LOKUTOR_API_KEY"

In ~/.cursor/mcp.json (or .cursor/mcp.json in a project; keep keys out of version control):

{
  "mcpServers": {
    "lokutor": {
      "url": "https://api.lokutor.com/mcp",
      "headers": { "Authorization": "Bearer sk_your_key" }
    }
  }
}

In .vscode/mcp.json. VS Code asks for the key once and stores it securely:

{
  "inputs": [
    { "type": "promptString", "id": "lokutor-key", "description": "Lokutor API key", "password": true }
  ],
  "servers": {
    "lokutor": {
      "type": "http",
      "url": "https://api.lokutor.com/mcp",
      "headers": { "Authorization": "Bearer ${input:lokutor-key}" }
    }
  }
}

Settings → Connectors → Add custom connector, URL https://api.lokutor.com/mcp. Custom connectors cannot send an API key header, so this gives the five discovery tools. Use Claude Code, Cursor or VS Code for synthesize_speech.

Protocol

Streamable HTTP, stateless. Every request is a POST of one JSON-RPC message (or a batch), answered with application/json. There is no session and no server-initiated stream, so GET returns 405. Protocol revisions 2025-11-25, 2025-06-18, 2025-03-26 and 2024-11-05 are supported.

curl -s https://api.lokutor.com/mcp \
  -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_languages","arguments":{}}}'

Lokutor Docs

Gives an assistant search_lokutor and a read-only view of every page and the OpenAPI spec, so it answers from the current docs instead of from memory. No key needed.

Claude Code
Cursor / VS Code
Other clients

claude mcp add --transport http lokutor-docs https://docs.lokutor.com/mcp

Add a Streamable HTTP server with the URL https://docs.lokutor.com/mcp. The page menu’s Copy MCP Server copies it.

Agent Skill: teaches a coding agent how to build with Lokutor
OpenAPI Spec: the machine-readable API contract
llms-full.txt: the complete docs in one file

Get Started

SDK

API Reference

Resources

Lokutor API

Tools

Add it

Protocol

Lokutor Docs

​Lokutor API

​Tools

​Add it

​Protocol

​Lokutor Docs

​Related

Lokutor API

Tools

Add it

Protocol

Lokutor Docs

Related