Getting started

MCP

The ZilAPI MCP server lets AI assistants list models, check your credit, run Noise Removal, Vocal Separation and Speech Enhancement jobs and fetch the results with your API key, over the Model Context Protocol.

npm package coming soon: @zilapi/mcp is not published on npm yet. Until it is, run the server from a checkout of its source, as shown below; the source repository is not public yet either, so ask support@zilapi.com for access. When the package is out, this page will show the one-line command that replaces the checkout path.

What it can do

MCP tools
Tool	Does
list_models	Lists live models, prices and limits. No key needed.
get_credits	Shows your available credit.
upload_file	Uploads one local audio file or public https URL and returns a ready file id.
denoise_audio	Removes noise (`zilapi/nr-4.1` by default).
separate_music	Splits a song into vocals and instrumental (`zilapi/vms-2.5`).
enhance_speech	Makes speech clearer (`zilapi/se-1.0`).
get_job, wait_for_job	Reads a job, or waits until it finishes and returns download links.
get_job_output	Refreshes download links; every output of a multi-output job.
cancel_job, list_jobs	Cancels a queued job; lists jobs.

Task tools take exactly one of a file_id, a local path or a url. There is also a clean-up-audio prompt that walks the assistant through credit, model choice, the job and the download links.

Build it from a checkout

You need Node.js 22 or later. Build once; every client below then runs dist/stdio.js from that folder.

Build the MCP server

git clone https://github.com/inverse-ai/zilapi-mcp.git
cd zilapi-mcp
npm ci && npm run build      # Node.js 22 or later
pwd                          # use this path below: <path>/dist/stdio.js

Claude Code

Create an API key in the dashboard under API keys. Start with a test key: test jobs are free.
Add the server, with your key as ZILAPI_API_KEY and the path to your checkout:

Add the server to Claude Code

claude mcp add zilapi --env ZILAPI_API_KEY=zil_test_... -- node /path/to/zilapi-mcp/dist/stdio.js

Run /mcp in Claude Code to check that zilapi is connected.

Cursor

Add a zilapi entry to ~/.cursor/mcp.json (or open Cursor Settings, then MCP, then add a new server), then check that it shows as enabled.

~/.cursor/mcp.json

{
  "mcpServers": {
    "zilapi": {
      "command": "node",
      "args": ["/path/to/zilapi-mcp/dist/stdio.js"],
      "env": { "ZILAPI_API_KEY": "zil_test_..." }
    }
  }
}

Claude Desktop

Open Settings, then Developer, then Edit Config. Add the same mcpServers entry to claude_desktop_config.json, then restart Claude Desktop.

claude_desktop_config.json

{
  "mcpServers": {
    "zilapi": {
      "command": "node",
      "args": ["/path/to/zilapi-mcp/dist/stdio.js"],
      "env": { "ZILAPI_API_KEY": "zil_test_..." }
    }
  }
}

Settings

MCP server environment
Variable	Default	Meaning
ZILAPI_API_KEY	unset	Your API key. Listing models works without one. (The MCP server reads `ZILAPI_API_KEY`; the docs' examples use `ZILAPI_KEY`.)
ZILAPI_BASE_URL	https://api.zilapi.com	The API to talk to.

Keys and costs

The MCP server uses an ordinary API key, so everything an assistant does is charged and limited like any other request. Live jobs spend credit.
Use a test key while you try it out: test jobs are free and return a sample result.
A key is a secret: keep it in the client's config or environment, never paste it into a chat. The server never logs or echoes it.