MCP FAQ – Podscribe

What is MCP

The Model Context Protocol (MCP) is an open standard that lets an AI app (Claude, ChatGPT, Cursor, and others) call tools on a remote server.

Podscribe hosts an MCP server. After you connect, your AI client can ask Podscribe questions about your podcast and audio advertising data — campaigns, attribution, audience, shows, publishers, creatives, and planning — without leaving that client.

The Podscribe MCP server is remote HTTP (Streamable HTTP). Production URL: https://backend.podscribe.ai/api/public/mcp. You can also copy it from Settings → MCP. Nothing runs on your laptop except the AI client itself.

Why you may need MCP

Use Podscribe MCP when you already work in an AI client and want Podscribe answers there:

Ask campaign, attribution, or research questions from Claude, Cursor, ChatGPT, or another MCP client.

Keep using the same AI Chat capabilities you already have in the Podscribe app, from the tools you use every day.

Leave a trail in Podscribe: every MCP question becomes a chat you can open in the app.

Some MCP clients support scheduling. You can create agents in that client that run prompts on a schedule - for example a weekly campaign recap - without opening Podscribe each time.

If you only use AI Chat inside Podscribe, you do not need MCP. MCP is for bringing that same assistant into other apps.

What tools we have now

Podscribe MCP exposes one tool:

Tool	What it does
`ask_podscribe`	Ask Podscribe a question in plain language. Same capability as AI Chat in the Podscribe app.

ask_podscribe answers from the data your Podscribe user can already see: campaign reporting and attribution, audience size and demographics, show and publisher research, creatives and airchecks, and planning.

Write one self-contained question. Include the date range, advertiser, campaign, or show names the answer needs. A single call can take a few minutes — Podscribe runs the same work AI Chat runs in the app.

The reply is markdown. It also includes a link to open that answer in Podscribe.

More tools may be added later.

How to connect

Create an MCP key in Podscribe

You need a Podscribe MCP key. The key is tied to your user. The server answers with the same access you have in the app.

Open Settings → MCP (direct link).

Copy the MCP server URL shown at the top of the page. Production URL is https://backend.podscribe.ai/api/public/mcp.

Click Create MCP Key. Optionally name it after the client (for example Claude Desktop or Cursor). If you leave the name empty, Podscribe names it from the creation time.

Copy the key immediately. This is the only time Podscribe shows it. It cannot be recovered later. Treat it like a password.

Paste the URL and key into your MCP client. See Authorization types and the client accordion below.

You can have up to 5 active keys. Disable one to create another. Disabling a key disconnects every client that uses it. That cannot be undone — create a new key if you still need access.

If you do not see an MCP tab in Settings, email support@podscribe.com.

Authorization types

Type	Status	How to use it
MCP key (Bearer token)	Available now	Create a key in Settings → MCP. Send it on every request as `Authorization: Bearer YOUR_MCP_KEY`.
OAuth	In progress	Sign in through your MCP client, no key to paste. We are building this as an additional way to connect.

Your client must:

Use the MCP URL: https://backend.podscribe.ai/api/public/mcp (also shown in Settings → MCP)

Send header Authorization: Bearer YOUR_MCP_KEY

Use Streamable HTTP (remote HTTP). Not stdio, not a local install.

Do not put the key in the URL. Do not commit it to git. If a key leaks, disable it in Settings → MCP and create a new one.

MCP clients

Copy the MCP URL (https://backend.podscribe.ai/api/public/mcp) and key from Settings → MCP first, then open the accordion for your client.

ChatGPT

ChatGPT connects to remote MCP servers from Developer mode (custom connectors / MCP apps). Header (Bearer) auth works.

Official docs:

ChatGPT Developer mode

Connect from ChatGPT

Developer mode and MCP apps (Help Center)

When ChatGPT asks for a server URL, use https://backend.podscribe.ai/api/public/mcp. Add an Authorization header:

Authorization: Bearer YOUR_MCP_KEY

Follow the official ChatGPT docs for Developer mode, creating the connector, and enabling it in a chat.

Claude Desktop

Two ways to connect Claude Desktop:

Claude CLI (recommended with an MCP key). Add the server from the terminal; Claude Desktop can use servers registered that way. See the Claude CLI accordion and Connect Claude Code to tools via MCP.

Remote custom connector in Claude. Official docs:

Custom connectors using remote MCP (Claude Help Center)

Connect to remote MCP servers

When Claude asks for the server URL, use https://backend.podscribe.ai/api/public/mcp. If the UI offers a custom header, send Authorization: Bearer YOUR_MCP_KEY.

Claude CLI

Claude CLI (Claude Code) can add a remote HTTP MCP server and send a Bearer header.

Official docs: Connect Claude Code to tools via MCP

Example (replace YOUR_MCP_KEY with the key from Settings → MCP):

claude mcp add --transport http podscribe https://backend.podscribe.ai/api/public/mcp \
  --header "Authorization: Bearer YOUR_MCP_KEY"

Follow the official docs for scopes (local vs user), listing servers, and removing a server. Claude Desktop can use a server you add this way.

Cursor

Cursor reads MCP servers from mcp.json (project .cursor/mcp.json or user ~/.cursor/mcp.json) and from Cursor Settings → MCP.

Official docs:

Cursor MCP

Install MCP servers in Cursor

Example remote server (replace YOUR_MCP_KEY with the key from Settings → MCP):

{
  "mcpServers": {
    "podscribe": {
      "url": "https://backend.podscribe.ai/api/public/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_MCP_KEY"
      }
    }
  }
}

Prefer environment-variable interpolation for the key if your Cursor version supports it — see the official Cursor docs. Do not commit the raw key.

After saving, reload MCP (or restart Cursor) and confirm ask_podscribe is listed.

Other MCP clients

Any client that speaks MCP over Streamable HTTP and can send a custom Authorization header can connect.

Use:

URL: https://backend.podscribe.ai/api/public/mcp

Header: Authorization: Bearer YOUR_MCP_KEY

Tool: ask_podscribe

Client directories and protocol docs:

MCP clients

MCP introduction

If the client only supports OAuth for remote servers, wait for Podscribe OAuth or pick a client that supports Bearer tokens.

Problems and support

If something fails:

Confirm the MCP tab shows the key as active (not disabled).

Confirm the client URL is https://backend.podscribe.ai/api/public/mcp.

Confirm the header is Authorization: Bearer … with the full key, no extra quotes inside the token.

Create a new key if you closed the create dialog without copying — the old secret cannot be shown again.

Check Podscribe chat history. A successful ask_podscribe call always creates a chat there, even if the client UI looks stuck.

For anything else, email support@podscribe.com. Include the MCP client name (ChatGPT, Claude Desktop, Claude CLI, Cursor, …) and roughly when you tried to connect. Do not send the MCP key in email.

🔄

OAuth is in progress. Today you connect with an MCP key sent as Authorization: Bearer. We are building OAuth so clients can sign in without pasting a key. Header auth already works in clients that support it, including ChatGPT.

💬

Every MCP question is saved in your Podscribe chat history, the same list as AI Chat in the app. Open Podscribe later to re-read the answer. MCP chats are marked with a cable icon.