Connect Podsquick to your agent

Podsquick is a remote MCP server over Streamable HTTP. Claude, ChatGPT and Gemini all speak it. Adding it takes a URL and an API key, and then you ask for a podcast in plain English.

The server URL is the same everywhere:

https://podsmith.pages.dev/mcp

You will also need an API key, which starts psk_. Mint one at /admin. Set a spend limit on it — the key is what pays for generation.

Claude

Settings → Connectors → Add custom connector. Paste the server URL, and put your API key in as a bearer token when prompted. Claude lists the available tools once it connects; if it does not, the URL is wrong or the key was rejected.

Available on Claude.ai (Pro, Max, Team and Enterprise) and in Claude Code via claude mcp add --transport http podsquick <url>.

ChatGPT

Settings → Apps & Connectors → enable Developer mode → Create. Paste the server URL, choose bearer-token authentication and give it your API key. The same listing works in Codex.

Developer mode is available on Plus, Pro, Business and Enterprise plans. It exists precisely so you can add a server that is not in the directory yet.

Gemini

Gemini Enterprise: add Podsquick as a custom MCP connector — Streamable HTTP, bearer authentication, the URL above. Your workspace administrator does this once and every user in the workspace gets the tools.

Gemini API: the SDKs take a remote MCP server directly. Point the client at the URL above with an Authorization: Bearer header.

The consumer Gemini app does not currently accept self-serve custom connectors.

Then just ask

Once it is connected, this is the whole interface:

Turn these three reports into a 10-minute podcast I can
listen to on the way in, and give me the transcript.

The agent calls generate_podcast, polls until it is done, and hands back the audio, a timestamped transcript, a citation manifest and show notes.

Two things worth telling your agent

Generation takes a few minutes and the agent has to poll for it. That is in the tool description, so most agents handle it — but if yours gives up early, tell it to keep checking get_job_status until the status settles.

Review before you spend on audio by asking for approval mode. The job stops after the script is written and audited, shows you the draft, and renders nothing until you say go.

No agent to hand?

The studio does the same thing in a browser, and the REST API does it from a script. Same pipeline, same artifacts.