Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@@useclypt/mcp-server Clip this podcast: https://feeds.simplecast.com/abc123"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Model Context Protocol server for Clypt. Lets AI agents (Claude Desktop, Cursor, Claude Code, any MCP-aware client) submit a podcast URL and receive clips, an optional trailer, show notes, a guest-share link, and the transcript.
Pure shim over the public REST API at https://useclypt.com/api/v1 — three tools, no clever middleware.
Tool | Wraps | Returns |
|
|
|
|
| Current job state. When |
|
| Cursor-paginated list of the org's recent jobs. |
The flow is asynchronous: submit_job returns immediately with a queued id; the agent polls get_job until terminal. Once a webhook is registered against your org, job.completed and job.failed events fire on every terminal transition — register via the public API at POST /v1/webhooks (no MCP tool for this yet; coming in a later minor).
Related MCP server: Podcast Index MCP
With the server configured in Claude Desktop, ask:
Use Clypt to clip this podcast and tell me when it's done: https://feeds.simplecast.com/abc123
Claude will call submit_job, get back a job id, poll get_job until terminal, and surface the clips + trailer + transcript inline. Wall-clock: ~2-5 min for audio/RSS, ~10-12 min for video + trailer.
npm install -g @useclypt/mcp-serverOr run without installing:
npx -y @useclypt/mcp-server~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"clypt": {
"command": "npx",
"args": ["-y", "@useclypt/mcp-server"],
"env": {
"CLYPT_API_KEY": "clk_live_..."
}
}
}
}Restart Claude Desktop. Then prompt: "Use Clypt to clip this podcast: ".
Same recipe — the binary is mcp-server (or run without global install via npx -y @useclypt/mcp-server) and the only required env var is CLYPT_API_KEY.
If your key starts with clk_test_ instead of clk_live_, every job returns a deterministic fixture instead of running the real pipeline. No transcription cost, no R2 storage cost, no Stripe charges. Useful for wiring up your agent before you commit to real submissions.
Fixture mapping:
source.type='video_url' (any URL) → success fixture with 3 clips + optional trailer
source.type='audio_url' / rss_feed_url (URL without "fail") → success fixture
Any URL containing "fail" → transcription_failed error fixture
source.type='youtube_url' → youtube_ingestion_failed error fixture
Malformed URL → invalid_source_url validation error at submit
Get a sandbox key in seconds at useclypt.com/developers/signup — no credit card, no waiting list. Sandbox keys (prefix clk_test_) return deterministic fixtures so you can wire up your agent for free before swapping in a clk_live_ key.
Variable | Required | Default | Purpose |
| yes | — | Bearer token for the Clypt API. |
| no |
| Override for staging or local development. Trailing slash is stripped. |
npm install
npm run build
npm test # vitest unit tests with mocked fetch
npm run dev # tsc --watchSmoke against the live API with the MCP inspector CLI:
npm run build
CLYPT_API_KEY=clk_test_xxx \
npx @modelcontextprotocol/inspector --cli node dist/index.js --method tools/listAPI reference: useclypt.com/developers/docs
Get a key: useclypt.com/developers/signup
About Clypt: useclypt.com
MIT — see LICENSE.
Fetch the current state of a previously submitted job. When status='complete', the returned envelope contains the output object with clips, optional trailer, guest_share_url, show_notes, and transcript. When status='failed', the envelope contains an error object with code + message. Call this repeatedly (every 30-60s is plenty — jobs take minutes, not seconds) until status is one of complete/failed.
Inputs: id (required).
List recent jobs for the authenticated organisation, newest first. Cursor-paginated via starting_after. Useful for inspecting recent submissions when you don't have the id to hand.
Inputs: limit, starting_after.
Submit a podcast for processing. Returns a job id immediately with status='queued'; the pipeline runs asynchronously — call get_job to poll. Typical wall-clock to terminal is ~2-5 min for audio_url/rss_feed_url, ~10-12 min for video_url with include_trailer=true. Source type 'youtube_url' is in beta — sandbox keys return a fixture failure; live keys reject it for now. For free deterministic testing, use a sandbox key (prefix clk_test_) — every submission returns canned fixture output ~10-70s later.
Inputs: source (required), options, idempotency_key.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 3 tools
Each tool has a clearly distinct purpose: submit creates a job, get retrieves a specific job, and list enumerates jobs. No overlap or ambiguity.
All tool names follow a consistent verb_noun pattern: submit_job, get_job, list_jobs. The slight pluralization in list_jobs is a minor and natural deviation.
Three tools is well-scoped for the server's purpose of managing asynchronous podcast processing jobs. Each tool earns its place with no redundancy.
The surface covers the core lifecycle: submit, poll, and list jobs. Missing cancel/delete is a minor gap but not essential for the stated async-processing workflow.
Podcast intelligence for agents: transcripts, clips, speaker diarization, mention tracking.
Turn podcasts and long videos into ranked, finished short clips via the hosted ClipZyn API.
Create, import, and publish Jellypod podcast episodes from your AI assistant.
Podcast search, metadata, chapters, and transcripts for AI agents — from $15/mo
