How to publish videos with an AI agent

Connect Taisly to an AI agent to prepare and execute reviewed publishing requests. This workflow uses remote MCP with OAuth, real destination IDs and an explicit status check after creation.

Connect once and discover the accounts

Add https://app.taisly.com/mcp in a client that supports remote MCP and OAuth, then complete browser authorisation. Use the client setup instructions or Agent Kit. Reuse an existing connection; remote OAuth MCP does not require a separate API key or CLI installation.

Ask the agent:

Check Taisly access with taisly_auth_status, then use taisly_platforms_list to show my accounts and exact IDs. Do not publish during setup.

An empty list means no social accounts are connected, not that reinstalling MCP will help. Connect the accounts you intend to use.

Provide a video the remote server can fetch

Remote MCP accepts a public video URL. A laptop path, private drive preview or expired signed URL may be inaccessible. Keep the URL available until processing finishes.

For a local-file workflow, follow the local upload instructions instead of sending the remote server a local path. Prepare the final export first.

Validate a complete request

Use this video URL: [my accessible video URL]. Prepare the caption “Try this squat stance check.” Use only the two account IDs I selected from your list. Run taisly_posts_validate. Show the exact accounts, caption and timing. Wait for my approval before taisly_posts_create.

The bracketed URL is a placeholder, not a working example file. For a future post, specify the full date and an explicit UTC offset, for example 2026-11-03T09:00:00+07:00. Replace the example with your intended future time.

Create after approval and retain the receipt

After the user approves the concrete request, the agent can call taisly_posts_create with confirmed: true. Save the returned historyId. A successful create response means accepted for processing, not successful delivery everywhere.

Ask:

Use taisly_posts_status for that history ID. Report each destination's status and any available post link. Do not create a replacement post while the existing attempt is pending.

A scheduled post remains future work until its release and destination processing complete.

Handle errors without repeating the whole request

If a destination fails, inspect its error and check the account itself. Reconnect or correct the file when needed. Request a separately approved retry only for the missing destination; do not resend successful accounts.

Use an agent when publishing is already part of an assisted workflow. Use the web scheduler for a reviewed calendar or Auto Repost for a stable source-driven route. The value is a controlled, inspectable process, not an unreviewed instruction to “post everywhere.”