Agents
Let an agent work with poddub for you, by signing in or with a key.
Where poddub works
poddub is one address, https://poddub.com/api/mcp, that each of these clients connects to, so there is nothing to install.
| Client | How it connects |
|---|---|
| claude.ai | Add poddub as a connector and sign in. It is then in Claude Code too, and, by claude.ai's documentation, in Claude Desktop and the Claude mobile apps under the same Claude account. Not tried yet beyond Claude Code. |
| Claude Code | Already there if you added poddub in claude.ai. Otherwise one command, a file shared with a project, the poddub plugin, or a key for automation. |
| Cursor | A key. Not tried yet. |
| VS Code | A key. Not tried yet. |
| ChatGPT | Add poddub as a custom MCP server and sign in. Not tried yet. |
Two ways to connect
- Signing in. You add poddub to the agent, and the agent sends you to poddub.com to sign in with your own poddub account. You choose the organization it works in and the most it may do, Member unless you choose more and never above your own role. Anyone in an organization can connect an agent this way. claude.ai, Claude Code and ChatGPT sign in.
- A key. An owner makes a key in the poddub app and gives it to the agent. A key is for work that runs without a person, such as a script or an automation, and for an agent that cannot sign in, such as Cursor or VS Code. See Keys.
Either way, the agent acts as its person and never does more than that person can.
What an agent can do
An agent you connect works with poddub the way you would in the app. It can read your teams, how many minutes and script generations are left, your sources, your episodes and their scripts. It can paste what shipped into a team's inbox and correct it. It can start an episode from what was pasted and shape its script: rewrite a line by hand, have poddub write a line or the whole script again, rename the episode. It can retry an episode that failed, and publish a ready episode to its team, or to the whole organization when your role allows it. Everything it does counts against the same allowance as the app, and it never does more than the role you allowed it, nor more than you can do yourself.
What an agent never does
An agent never:
- adds or removes people, sends invitations, changes roles, or manages teams;
- buys, changes or restores a subscription;
- deletes a source, an episode, an organization or an account;
- puts an episode on the public show or takes one back, unpublishes an episode, or narrows who hears it;
- accepts or withdraws your consent to the providers poddub works with;
- reads or replaces a private feed link, or sees who listens;
- creates, reads or revokes a key;
- reaches anything only poddub's own operators can do.
None of these has a tool, so no prompt can make an agent do them.
What ends a connection
A connection made by signing in stops at once:
- when you end it on poddub.com/connections, the page called Connected agents, where you see your own agents and an owner sees everyone's in their organization;
- when you leave or are removed from the organization;
- when the organization is deleted;
- when your poddub account is deleted.
A key stops in the same cases, and when it expires or is revoked: see Keys.
Approval
A person approves an episode, in the app. Approving spends the organization's minutes and starts the audio, so an agent stops when the script is ready and tells you so. An owner can allow agents to approve in their organization. It is off until they turn it on, and the app records who turned it on and when.
Where to start
- Connect claude.ai, Connect Claude Code and Connect ChatGPT: signing in.
- Connect Cursor and Connect VS Code: connecting with a key.
- Keys: what a key is, who makes it, and what stops it.
- An example: release notes to a published episode, step by step.
- Tools: every tool, what it takes and what it can be refused with.
- Rules and limits: what an agent works within.