PodLearn MCP Server
by blutrich
README.md
<div align="center">
# 🎧 PodLearn for Claude Code
### Stop listening. Start learning.
**The podcast & video knowledge engine, right inside Claude Code.**
Drop a link. Get a verbatim transcript, a searchable archive, and an AI lesson in seconds.
[](https://code.claude.com/docs/en/plugins)
[](https://podlearn.ai)
[](https://podlearn.ai)
[](./LICENSE)
</div>
---
Some of the best thinking on the planet is locked inside hours of audio you'll
never have time for. PodLearn unlocks it. Hand Claude any **podcast or YouTube
URL** and it transcribes the whole thing (yes, even long talks with no captions),
makes every word searchable, and distills the ideas into a structured lesson.
No tab-switching, no copy-paste, no "I'll get to it later."
One install adds the PodLearn **skill *and* MCP server** to Claude Code. The
skill handles the routing; the MCP does the work. You just talk.
## ✨ Why people use it
- ⏱️ **Hours → minutes.** A 2-hour interview becomes a transcript + lesson you skim in five.
- 🔎 **Find the exact moment.** Full-text search across every transcript, with citations. *"Where did they talk about pricing?"* jumps you straight there.
- ▶️ **YouTube that actually works.** Paste any video, captions or not. Conference talks, lectures, channel deep-dives, all transcribed natively.
- 📚 **Bulk a whole season.** `transcribe_feed` ingests an entire podcast feed in one shot, not episode-by-episode.
- 🧠 **Lessons, not just text.** Auto-generated key takeaways and structured notes you can act on.
- 🤖 **Agent-native.** It's a real MCP server, so every call composes with the rest of your Claude workflow (turn a lesson into a LinkedIn post, feed quotes into a doc, whatever).
## 💬 What you can ask
> *"What podcasts do you have about AI agents?"*
> *"Transcribe this and make me a lesson: ‹youtube or podcast URL›"*
> *"Find the best quotes about pricing from ‹episode›"*
> *"Transcribe the last 10 episodes of ‹podcast› and summarize the themes."*
> *"Turn that lesson into a LinkedIn post."*
## ⚡ Install (60 seconds)
> **1. Have a PodLearn account** (free, 2 credits to try) at **[podlearn.ai](https://podlearn.ai)**. No token needed: you sign in once in your browser.
**2. Add the marketplace** — in Claude Code (run this on its own, wait for confirmation):
```
/plugin marketplace add blutrich/podlearn-plugin
```
**3. Install the plugin** — then run:
```
/plugin install podlearn@podlearn
```
> ⚠️ Run them **one at a time** — they're two separate slash commands. Pasting
> both lines at once makes Claude treat the second as part of the first.
**4. Restart Claude Code** (or `claude reload`).
**5. Sign in.** Run `/mcp`, pick **podlearn**, choose **Authenticate**, and
approve PodLearn in the browser. Approve both read and write if you want
Claude to transcribe and save things for you (it still asks before spending
credits). Then start asking.
That's it. The plugin ships the skill **and** the MCP, so there's no separate
`claude mcp add` step and nothing to wire up by hand.
## 📦 What's inside
| | |
|---|---|
| **Skill** (`podlearn`) | Routes your intent through the MCP tools so you never think about which one to call. |
| **MCP server** | `search_podcasts` · `search_youtube` · `search_episodes` · `list_feed_episodes` · `transcribe_url` · `transcribe_feed` (bulk/season) · `get_transcription` · `search_transcription` · `generate_lesson` · `generate_linkedin_post` · and more. |
## 🔌 Just the MCP server (without the plugin)
The plugin already bundles the server, so most people don't need this. But if you
want the MCP on its own — in **Claude Desktop**, **Cursor**, **ChatGPT**, or any
other MCP client, or in Claude Code without the skill — add it directly.
**Claude app** (web, desktop, mobile): Settings → Connectors → Add custom
connector → paste `https://podlearn.ai/mcp` → Connect, and sign in.
**Claude Code:**
```bash
claude mcp add --transport http podlearn https://podlearn.ai/mcp
```
Then `/mcp` → **podlearn** → **Authenticate**.
**Codex:**
```bash
codex mcp add podlearn --url https://podlearn.ai/mcp
codex mcp login podlearn --scopes mcp:read,mcp:write
```
**Any MCP client** — point it at the Streamable HTTP endpoint and sign in with OAuth:
| Field | Value |
|---|---|
| Transport | Streamable HTTP (`http`) |
| URL | `https://podlearn.ai/mcp` |
| Auth | OAuth (sign in with your PodLearn account) |
Headless client that can't open a browser? Create a personal token in
[Settings → API Tokens](https://podlearn.ai/settings) and send it as the
`Authorization: Bearer …` header. Never put it in the URL: tokens in the
query string are ignored, and calls will hang waiting for a sign-in.
Full setup guide: [podlearn.ai/mcp-docs](https://podlearn.ai/mcp-docs).
## 🔄 Manage
```
/plugin marketplace update podlearn # pull the latest
/plugin uninstall podlearn@podlearn # remove the plugin
/plugin marketplace remove podlearn # remove the marketplace
```
## 🔐 Notes
- **Your account, your permissions.** The MCP signs in with OAuth; you choose
read-only or read + write in the browser. Auth error or "insufficient scope"?
Run `/mcp` → **podlearn** → re-authenticate. To get write access after a
read-only sign-in, remove and re-add the server first.
- **Upgrading from 1.0.x?** The old token prompt is gone. Update the
marketplace, reinstall, restart, then sign in with `/mcp`. You can revoke the
old `pl_mcp_…` token in [Settings → API Tokens](https://podlearn.ai/settings).
- **Transparent pricing.** Reading and searching transcripts already in your
library is free. Paid actions (`transcribe_url`, `transcribe_feed`,
`generate_lesson`) spend PodLearn credits, and the skill confirms the cost
before charging. Start with 2 free credits; top up or subscribe at
[podlearn.ai](https://podlearn.ai).
## 🌐 Also available
PodLearn runs as a web app at **[podlearn.ai](https://podlearn.ai)** and as an
MCP server for Claude Desktop, ChatGPT, Cursor, and any MCP-compatible client.
This plugin is the fastest way to use it from Claude Code.
## 📄 License
[AGPL-3.0](./LICENSE) — © 2026 Ofer Blutrich. Built on the
[PodLearn](https://podlearn.ai) platform.
This server cannot be deployed
Maintenance
ActivityMaintained
ResponsivenessUnresponsive