媒体管理 MCP 服务器

Requesting and adding media

request_media creates a request in Seerr and uses its approval workflow. TV requests default to season 1; specify seasons to request others.

You can also add media directly with radarr_add_movie or sonarr_add_series. These bypass Seerr. They require a quality profile and root folder from radarr_get_options or sonarr_get_options, and don't start a search automatically.

IDs depend on the operation: Seerr requests and Radarr additions use TMDB IDs; Sonarr additions use TVDB IDs. Once a title is in your library, monitoring and search tools use its Sonarr or Radarr ID. Episode tools use Sonarr episode IDs and check that they belong to the selected series.

Sonarr monitoring changes only the selected seasons or episodes. Season and episode changes require separate calls. All of this works in read-only mode except request_media and the direct additions.

Seerr request history

"Who asked for this?" is a request-log question, not a library one:

• list_requests pages the request log newest-first (page_number, limit up to 100) and carries the requester on every entry. requested_by is who asked; modified_by is whoever last acted, usually the approving admin, and the two routinely differ. Narrow with requested_by (Seerr user id) and media_type, or filter.
• get_media_requests answers it for one title by TMDB id, oldest first, so the first entry is the original requester. Use it instead of scanning the log when the title is known.

Both are read-only and need no writes. Two upstream quirks shape them: /api/v1/media deliberately omits requests, and request?mediaId= is silently ignored (it returns the whole log), so per-title history only comes from the title's own detail endpoint. Seerr also answers 500, not 404, for a TMDB id it has never seen, which these tools report as in_seerr: false rather than an error. Filters, sorts and media types are validated locally because Seerr rejects a bad value with 400 but silently ignores an unknown requestedBy.

Paging is reported as native from Seerr's pageInfo; a title the requester list does not match returns an empty page rather than an error.

Fixing and removing requests

A request that never reached Sonarr or Radarr sits in the log forever without ever downloading. The signal is media.externalServiceId being empty on a request that is otherwise approved. retry_request is Seerr's own repair for it: it re-runs the push and reports pushed_to_arr from the result, instead of creating a second request. set_request_status approves or declines one request (approve or decline); approval is the decision that lets Seerr push to the arr, not a bookkeeping flag.

Removal comes in two deliberately separate operations, because they delete different things:

• remove_request clears ONE log entry. The title's media record survives, so it can be requested again later. A request Seerr has already forgotten reports removed: false instead of failing.
• remove_media deletes ONE Seerr media item, discarding its availability status, request history and availability state. It does not delete the file on disk and does not remove the title from Sonarr or Radarr.

A media ID is neither a TMDB ID nor a request ID. Seerr also accepts any media ID with a success response, even one it never had, so remove_media reports removed (the delete was accepted) separately from verified (a record is known to have existed). Neither tool can be used to delete files.

NZBHydra downloads

Upcoming episodes and movie releases

sonarr_get_calendar and radarr_get_calendar take start and end as ISO dates or timestamps, with a maximum window of 90 days. The Sonarr calendar includes series titles. Both calendars cover the library, not a single title; results include hasFile to show whether the episode or movie is already downloaded.

For one show's upcoming episodes, use sonarr_get_next_up with its Sonarr series_id. It lists monitored episodes in air-date order, starting from now. Set since to use a different date, or include_unmonitored=true to include episodes you aren't monitoring.

Searches and downloads

radarr_search_movie and sonarr_search_episodes start a search in Radarr or Sonarr, which may download a matching release. Sonarr accepts up to 100 episode IDs from one series per call. The response includes a command ID; check it with radarr_get_command or sonarr_get_command to see whether the search finished. A finished search doesn't mean the download or import is complete.

To choose a release yourself, use radarr_get_releases or sonarr_get_releases, then the matching grab_release tool. Sonarr requires an episode or season. These searches use indexer quota, and grabs refuse releases that are stale or rejected by the service.

Downloader tools require backend="nzbget" or backend="sabnzbd". Pause and resume take either a job_id from that downloader's queue or whole_queue=true. Omitting both is an error. Sonarr/Radarr queue and history entries include downloadId for matching a download to its job. NZBGet IDs are positive integer strings; SABnzbd IDs are opaque strings.

downloads_retry retries one failed history item. NZBGet downloads it again and may replace the failed partial files.

sonarr_remove_queue_item clears one Sonarr queue entry by its queue item_id. By default it leaves the downloader's files alone and doesn't blocklist the release. remove_from_client=true also requests removal from the downloader and may delete its files; blocklist=true marks the release failed. Queue removal requires Sonarr write permission.

Plex watch history (Tautulli)

Tautulli tools answer what has actually been watched, rather than what is available. They are read-only: there is no tool to delete history, terminate a stream, edit a library or change Tautulli settings, and none needs write permission.

Free text from Tautulli (titles, user names, status messages, summaries) is truncated to 512 characters; ids, timestamps and other numeric values are returned intact.

IDs come from Plex, not Sonarr/Radarr: history and stats use a Plex rating_key, where a show is the grandparent key and an episode or movie is its own. Use tautulli_search to turn a title into a rating key, or tautulli_get_metadata once you have one. tautulli_get_library_media_info needs a section_id from tautulli_get_libraries.

tautulli_get_activity shows only streams playing right now, so an empty result means nobody is streaming at this moment, not that an item was never watched; use tautulli_get_item_watch_stats for that. Tautulli's own users (as it tracks viewing) feed tautulli_get_user_watch_stats, while tautulli_get_users lists the Plex accounts that can reach the server.

Older-watched questions split two ways. tautulli_get_last_played returns the single most recent play with its timestamp and who played it, and needs media_type because Tautulli records each play against the actual file: a movie or episode matches on its own rating key, a show matches through its episodes' grandparent key, and a season through the parent key. The tool picks that column for you, so pass media_type="show" for a series and media_type="movie" for a film. played: false in the reply means no plays are recorded, which is the correct answer for an untouched title. For a list of past plays instead of just the latest one, use tautulli_get_history.

tautulli_get_item_watch_stats and tautulli_get_user_watch_stats return per-window rows (watch_time) plus the users or players involved. Their total_plays is all-time only, and is null when you ask for windows that exclude all time (query_days without 0), because Tautulli's windows nest (1 day sits inside 7, which sits inside 30) and cannot be added up.

Downloading from NZBHydra2

nzbhydra_search searches your indexers through Hydra. nzbhydra_get_capabilities lists the supported filters and categories.

To send a result to a downloader, call nzbhydra_download_result with its result_id, a backend and a configured downloader category. ArrChestra fetches the NZB and submits it to that downloader. The category's existing post-processing settings still apply.

This bypasses Sonarr/Radarr release selection. It doesn't fulfill a Seerr request or guarantee that the download will be imported into your library.

If NZB retrieval redirects to an indexer, add that indexer's origin to NZBHYDRA_ALLOWED_REDIRECT_ORIGINS. ArrChestra doesn't forward Hydra credentials or cookies to a different origin, follow HTTPS-to-HTTP redirects, or follow same-origin redirects outside the configured service path.

The Hydra submission tool appears only when a downloader is configured. Downloader calls always require an explicit backend.

Your MCP client's tools/list response includes the argument schemas. Most lists default to 25 results, up to 100 per page. Sonarr/Radarr queue, history and Sonarr missing episodes use page_number; other paged lists use offset. APIs that return a whole list are paged locally.