| name | studio |
| description | Operate a self-hosted Studio deployment — write, publish, schedule and edit posts and short-form video, read publication and audience analytics, and diagnose a post that did not reach a platform. Use whenever the request is about publishing, a channel, a draft, a scheduled post, delivery to Telegram/X/Threads/Instagram/YouTube, or why something did not go out. |
Studio
The studio MCP server is the whole interface to one deployment. There is no repository, database
or SSH here: if a task cannot be done with a tool, it cannot be done. Never ask for the server, the
container or the token.
Deployments differ — some publish a website and two languages, some are video only. Call
studio_capabilities before choosing a command, and read tools/list for the actual surface. Those
describe the deployment you are connected to; this file does not.
Publish or prepare
Take the wording literally, and when it is ambiguous, prepare.
- "Post it", "publish now", "send it" — publish: create,
studio_post_validate, then
studio_post_publish. Video: studio_video_preflight, then studio_video_publish.
- "Let's prepare a post", "draft this", "get it ready" — create and stop at the draft. Report
studio_post_preview and wait. The draft is already visible for approval in the deployment's own
Telegram bot and Command Center. Do not publish it later without being told to.
- A time — "tomorrow at nine" — is
studio_post_schedule or studio_video_schedule, not publish.
Media is attached to a draft, not passed to publish: studio_post_attach_media after the file is
uploaded to the deployment. A post with media publishes only once the media is attached.
Reaching an audience
studio_post_publish, studio_video_publish, ops_retry, ops_edit, ops_delete and
ops_use_other_media change what other people see. Run them only when asked for that specific
thing, once, and report what actually happened rather than what was intended. Publishing twice
because a call looked like it failed is worse than reporting the uncertainty.
Diagnosing
Start with ops_status — the guide is a CLI command, not a tool an agent can call — then:
ops_audit — failed jobs, stuck targets and inconsistencies across the text and video pipelines.ops_recent — the last posts, their targets, and the targets each one is missing. This is the
whole diagnosis for "post X did not go to Y".ops_verify — fetches a published target and reports whether it is really live.ops_find with a fragment of the text resolves an older post's ref. Refs are post:N.
Read the tool output before reasoning about causes, and never retry a target blindly: ops_recent
first, then one ops_retry for the target actually missing.
Analytics
studio_analytics_dashboard is the overview; studio_analytics_post_metrics,
studio_analytics_video_metrics and studio_analytics_audience are the detail. Audience numbers
come from live platform APIs and can legitimately be absent — that is not a bug to chase. Metrics
are sampled over time, so a post published minutes ago having no numbers is expected.
Language
Post text, captions and titles are written in the language of the channel they publish to.
Everything written to the operator — reports, questions, summaries — is in the operator's language.
