Connect your agent · Task 01 · 5 min
Connect your agent
Create an agent token in your channel, pick a door (MCP, the command line or the API) and work beside the app.
Your own AI agent can work in your Syndcast channel. It reaches Syndcast through one of three doors: MCP, the command line or the HTTP API. The MCP door uses a token you create in your channel's settings. This page covers the token and helps you pick a door.
Before you start
You need to be the channel's owner. Only the owner can manage agent tokens.
Step 1 of 6
Open Agent Access
Open your channel and press Channel Settings. Open the Agents tab. The Agent Access card lists your tokens. If you have none, it says "No agent connected yet". This is not the channel's Agent tab, where you brief Syndcast's own agent (Work with agents); tokens for your own agent live only here.
Step 2 of 6
Create a token
Press New Token. The Create agent token dialog opens.
- Token name: a name that tells you where the token is used, such as "Claude Code on my laptop".
- Role: Editor, Viewer or Owner. The agent gets the access of that role.
- Autonomy: Assist, Standard or Autopilot. The dialog says what each one allows. On Autopilot, set a Credit budget: the credits this token may spend before it asks again.
- Project scope (optional): pin the token to one project, or leave it on Whole channel.
- Content scope (optional): narrow the token to one piece of content in that project.
- What this agent may do: switch pages and provider preferences on or off.
Press Create. The dialog shows Token created.
Copy the token now
The token is never shown again. Copy it before you press Done.
Next, connect it: see Connect an MCP client.
Step 3 of 6
What a connected agent can do
Your agent works with the role, autonomy, scope and switches of its token. With the Owner or Editor role it can also:
- Brief Syndcast's own agents. Each task it starts pays for its model's tokens as it runs. The balance must hold a credit for it to start, and that credit is not spent.
- Make helper tokens for other agents. A helper never has more access than its parent and lasts 24 hours at most.
A token you create works for one year. See How your agent and Syndcast's agents work together.
Step 4 of 6
Change or revoke a token
Each token in Active tokens shows its role and scope. You can change what it may do and change its credit limit from its row.
To cut a token off, press its revoke button. The dialog asks you to revoke the token by name; press Revoke key. This cannot be undone. Revoked tokens move to Revoked tokens. Revoking a token also revokes its helpers, but tasks it gave to Syndcast's agents keep running until you stop them in the channel's Agent tab.
Step 5 of 6
Pick a door
| Door | Use it when | Page |
|---|---|---|
| MCP | You work in an agent client, such as Claude Code or Cursor. | Connect an MCP client |
| The command line | You work in a terminal or write a script. | Use the command line |
| The API | You write your own code. | Use the API |
All three doors reach the same catalog of operations, with the same key, reach, credits and limits behind them.
Step 6 of 6
Working beside the app
An external agent works beside you, not on its own. Today, through every door:
- You see its work in the app as it happens.
- When a request needs your yes, it waits in your channel's Agent tab. You approve or decline it there, and the agent picks up your answer within seconds.
- MCP and the command line wait for your answer. The API does not wait: it answers
approval_required, and the caller reads the open request withconfirmations.list.
Coming: chatting with an external agent from the browser, and addressing it from the editor. These do not work yet.