> ## Documentation Index
> Fetch the complete documentation index at: https://docs.audiopod.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# MCP Server
> Connect any agent to AudioPod via Model Context Protocol — music generation, TTS, voice cloning, transcription, and more from one endpoint.
**What is MCP?** [Model Context Protocol](https://modelcontextprotocol.io) is an open standard that lets AI assistants (Claude Desktop, Cursor, Continue, Cline, Goose, Codex…) call external tools through a single transport. AudioPod publishes a Streamable-HTTP MCP server that exposes ten audio AI tools.
## Endpoint
| Field | Value |
| --------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| URL | `https://mcp.audiopod.ai` |
| Transport | `streamable-http` (MCP 2025-06-18) |
| Auth | `X-API-Key: ap_*` (from your dashboard) **or** `Authorization: Bearer ` |
| Discovery | [`/.well-known/mcp/server-card.json`](https://audiopod.ai/.well-known/mcp/server-card.json) and [`/.well-known/oauth-protected-resource`](https://audiopod.ai/.well-known/oauth-protected-resource) |
Need a key? [Create one in your dashboard](https://www.audiopod.ai/dashboard/account/api-keys). Free tier credits unlock the full tool surface — no card required.
***
## Tools exposed
The server advertises ten tools. Schemas are returned by `tools/list` — these are the names and one-liners:
| Tool | Description | Required scope |
| ------------------- | ------------------------------------------------------------------------------------------------------------------------- | ------------------ |
| `generate_music` | Songs, instrumentals, rap, vocal stems from a text prompt. | `music:generate` |
| `text_to_speech` | TTS in 85+ languages with 500+ voices and custom clones. | `voice:synthesize` |
| `clone_voice` | Clone a voice from a 5–30s reference clip. | `voice:clone` |
| `separate_stems` | Up to 16 stems (vocals, drums, bass, guitar, piano, other…). Two-stem mode returns a vocals/instrumental (karaoke) split. | `stems:separate` |
| `transcribe_audio` | Word-level timestamps, speaker diarization, SRT/VTT/JSON. | `transcribe` |
| `denoise_audio` | Remove background noise while preserving voice character. | `audio:write` |
| `convert_media` | MP3 / WAV / FLAC / OGG / M4A / MP4 / MOV conversion. | `audio:write` |
| `separate_speakers` | Isolate each speaker into a separate track from multi-speaker audio. | `audio:write` |
| `change_voice` | Convert a recording to a different target voice (voice-to-voice). | `voice:synthesize` |
| `check_job_status` | Poll the status and result of a long-running job. | *none (free)* |
Scopes are enforced when authenticating with an API key. JWT-based auth implies full user privileges.
***
## Setup
### One-line install (recommended)
Most modern MCP hosts now register an HTTP MCP server with a single command — no manual JSON edit, no restart needed beyond the host's normal config reload. Pick the one that matches your tool:
```bash theme={null}
claude mcp add --transport http audiopod https://mcp.audiopod.ai \
--header "X-API-Key: ap_YOUR_KEY" --scope user
```
The `--scope user` flag installs at the user level so every Claude Code project picks up the server. Run `claude mcp list` to verify, then `/mcp` inside the TUI to see the ten tools.
```bash theme={null}
codex mcp add audiopod --url https://mcp.audiopod.ai \
--header "X-API-Key: ap_YOUR_KEY"
```
Verify with `codex mcp list`. Inside the Codex TUI, `/mcp` shows the connected servers and their tools.
For any MCP client that only speaks stdio (older Claude Desktop builds, Continue, Cline before they ship native HTTP support, etc.), the [`mcp-remote`](https://www.npmjs.com/package/mcp-remote) bridge spawns a stdio process that proxies to our HTTP endpoint:
```bash theme={null}
npx -y mcp-remote https://mcp.audiopod.ai \
--header "X-API-Key: ap_YOUR_KEY"
```
Wire it in as a stdio server in your client's config. Useful when manual JSON is too fiddly.
### Manual config (for desktop apps without a CLI)
Edit `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%\Claude\claude_desktop_config.json` (Windows):
```json theme={null}
{
"mcpServers": {
"audiopod": {
"url": "https://mcp.audiopod.ai",
"headers": {
"X-API-Key": "ap_YOUR_KEY"
}
}
}
}
```
Restart Claude Desktop. Try: *"Generate a 30-second lo-fi instrumental for studying."*
Edit `~/.cursor/mcp.json`:
```json theme={null}
{
"mcpServers": {
"audiopod": {
"url": "https://mcp.audiopod.ai",
"headers": { "X-API-Key": "ap_YOUR_KEY" }
}
}
}
```
Reload Cursor. Try: *"Use audiopod to generate a 60s upbeat intro track and download it."*
Edit `~/.continue/config.json`:
```json theme={null}
{
"experimental": {
"modelContextProtocolServers": [
{
"transport": {
"type": "streamable-http",
"url": "https://mcp.audiopod.ai",
"headers": { "X-API-Key": "ap_YOUR_KEY" }
}
}
]
}
}
```
Edit `~/.config/cline/cline_mcp_settings.json`:
```json theme={null}
{
"mcpServers": {
"audiopod": {
"url": "https://mcp.audiopod.ai",
"transport": "streamable-http",
"headers": { "X-API-Key": "ap_YOUR_KEY" }
}
}
}
```
Run `goose configure` and add a Streamable-HTTP extension:
```
Type: streamable-http
URL: https://mcp.audiopod.ai
Header: X-API-Key=ap_YOUR_KEY
```
Probe the server directly to confirm your key works:
```bash theme={null}
curl https://mcp.audiopod.ai \
-H "X-API-Key: ap_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/list"
}'
```
***
## JSON-RPC examples
### Initialize
```json theme={null}
POST https://mcp.audiopod.ai
{
"jsonrpc": "2.0",
"id": 1,
"method": "initialize",
"params": {
"protocolVersion": "2025-06-18",
"clientInfo": { "name": "my-agent", "version": "0.1.0" }
}
}
```
### List tools
```json theme={null}
{
"jsonrpc": "2.0",
"id": 2,
"method": "tools/list"
}
```
### Call a tool
```json theme={null}
{
"jsonrpc": "2.0",
"id": 3,
"method": "tools/call",
"params": {
"name": "generate_music",
"arguments": {
"prompt": "lo-fi rainy 90 BPM, soft pads",
"duration": 60,
"task_type": "text2music"
}
}
}
```
The response includes a `structuredContent.job_id` — long-running tools return a job id immediately. Poll the matching `/api/v1//jobs/` endpoint, or call `tools/call` again with the `jobs` lookup tool wired into your CLI.
***
## Auth flow
AudioPod publishes RFC 9728 OAuth Protected Resource Metadata so clients can discover how to authenticate:
```bash theme={null}
curl https://audiopod.ai/.well-known/oauth-protected-resource
```
Returns the issuer, supported scopes, and the bearer-token pattern. Most agent clients use a simple `X-API-Key: ap_*` header — get your key in the [dashboard](https://www.audiopod.ai/dashboard/account/api-keys).
***
## Troubleshooting
1. Re-check your config file path — most agents won't reload after edits without a restart.
2. Confirm the URL is exactly `https://mcp.audiopod.ai` (no trailing slash, no `/sse`).
3. Hit the discovery URL: `curl https://mcp.audiopod.ai` returns a small JSON descriptor when the proxy is healthy.
* Verify your key starts with `ap_`.
* Check the key's scopes — the [dashboard](https://www.audiopod.ai/dashboard/account/api-keys) shows the per-tool scopes granted.
* For wildcard access use scope `*`.
Music and stem extraction return job ids immediately and resolve over the next 30–180 seconds. Most clients display the partial result; if yours doesn't, fall back to the [REST API](/api-reference/music/jobs) to poll for completion.
***
## Next steps
Same ten tools, terminal-first.
Single-pane overview of every developer surface.
Free credits, no card required.
Model Context Protocol reference.