OpenPostUpload media
Reuse workspace media or upload a public URL or local file.
An MCP assistant can reuse media that is already in the selected OpenPost workspace. Ask it to list recent media, choose the correct item, and attach the returned media ID to a Post.
Upload from a public URL
With mcp:full, an assistant can use upload_media_from_url for a public HTTP or HTTPS URL. OpenPost rejects URLs that resolve to private, loopback, link-local, multicast, or other local network addresses.
OpenPost downloads the file and applies the normal media limits, workspace permissions, quota, storage, deduplication, and analysis checks. A URL upload creates a workspace media item that the assistant can attach to a Post.
Use this path only when the media is already available at a public URL. Do not create a temporary public link for a private file when your client supports local uploads.
Upload a local file
A compatible MCP Apps client can open the file picker through render_local_media_upload with read access. Creating the app-only upload ticket and uploading the file require mcp:full.
The file travels from the app to OpenPost. Its bytes and upload credential do not enter the model's conversation context. The one-use upload ticket expires after ten minutes and is bound to the authenticated actor and selected workspace.
The local picker works only in clients that support MCP Apps UI and app-only tool calls. If your client can read local files but cannot show the picker, it can use upload_media_base64 with mcp:full and workspace editor access. Pass workspace_id, filename, and content_base64; mime_type and alt_text are optional. Use standard base64 or a data:<mime-type>;base64, URL. The file must be 8 MiB or smaller, and the complete MCP request must fit within 12 MiB. This path sends the encoded file through the assistant's tool arguments. OpenPost applies the same validation, quota, storage, deduplication, and analysis checks. For larger files, upload in OpenPost first and ask the assistant to reuse the media item.
Check the result
After an upload, ask the assistant to confirm the media item's ID, type, dimensions or duration, and analysis state. Add alt text before publishing an image when the destination supports it.
Fit an image to a destination
With mcp:full and workspace editor access, ask the assistant to call prepare_media_image with media_id, account_id, and the destination's explicit output_profile. Pass destination settings when they affect its limits, and optionally alt_text for the copy.
The operation returns a separate, fully processed image with status: "ready". It uses connected-account limits to pad, downscale, or re-encode the image. It retains the original and rejects animated images instead of dropping frames. Ask the assistant to attach the returned media ID explicitly, then validate the complete post. Preparation never replaces an existing attachment or publishes a post.
Edit an open project
Sign in to the workspace and open an Image Editor or Video Editor project. Connect Codex or Claude Code with editing access, then ask the assistant to find that open project. No OpenPost desktop installation is needed.
For example: "In my open video, make the title larger, use a favorite text style that fits, and show me the result." The assistant can inspect the actual timeline or layers, reuse available templates, apply supported edits and render a preview. Changes appear live in the editor and use its normal history.
Keep the browser project open while the assistant edits or exports. Local sources stay in the browser, but sampled frames, short audio previews, transcripts and scene captions may reach your assistant's model provider. Missing sources or incomplete analysis remain visible limitations.
Use "remember this choice" or "save this style" only when you want a persistent preference. Ordinary corrections affect the current edit. The built-in Assistant offers a style selector and Preferences for changing or forgetting rules and turning choice learning off. Choosing a style affects future requests and does not restyle existing content.
The built-in Assistant on Hosted requires a signed-in editor on an active paid plan. External MCP uses ordinary workspace permissions. Self-hosted Assistant uses the instance's configured AI provider. Unattended editing with the browser closed is not supported.