Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
UOINK_MCP_TOOLSETNoSet to `core` to expose the previous 35 tools instead of all 91 registry tools. Cursor and other clients with tool caps may need this setting. Restart the connection after changing it.
UOINK_USAGE_SHARINGNoSet to `0` to force anonymous usage sharing off.

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

Uoink turns videos, playlists, and podcast episodes into local AI-ready corpora. Use the tools to capture, transcribe, publish, search, inspect, and analyze saved uoinks. ffmpeg and ffprobe are required. Install them: sudo apt-get install ffmpeg (Debian/Ubuntu), sudo dnf install ffmpeg (Fedora, RPM Fusion), sudo pacman -S ffmpeg (Arch), apk add ffmpeg (Alpine). (media_requires_ffmpeg)

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": false
}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}
experimental
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
uoink_videoA

Capture one YouTube video into a saved Markdown corpus with screenshots. Use for a single video; use uoink_playlist for a batch or uoink_url for other supported video platforms and long-video modes. Runs synchronously and needs the local media runtime. Returns the saved corpus and file paths; failures can leave partial capture files. Recapturing a source can replace its saved corpus and metadata.

uoink_playlistA

Start a background capture job for the videos in a YouTube playlist. Use for a one-time batch; use add_monitored_playlist for ongoing metadata monitoring or uoink_video for one video. Returns a job_id for get_job_status and cancel_job. Requires the local media runtime; the job can produce partial outputs before completion or cancellation. Recapturing a source can replace its saved corpus and metadata.

get_job_statusA

Read progress, per-item outcomes and errors for a saved capture or transcription job. Use after uoink_playlist or transcribe_podcast_episode returns a job_id; use cancel_job to request cancellation. Returns the current job object, or an error for an invalid or unknown identifier.

cancel_jobA

Request cancellation of a capture or transcription job while keeping partial outputs on disk. Use when the user wants work stopped; use get_job_status to inspect progress without cancelling. Returns the updated job object or an error for an unknown identifier. Cancellation does not delete captures already saved.

list_recent_uoinksA

List the newest saved captures from the local index, with their slugs, titles, folders and capture times. Use to browse recent additions without a query; use search_uoinks for matching text or search_library for bounded citation-ready retrieval. Returns up to the requested number of items under the legacy yoinks key.

search_uoinksA

Find whole saved items by text, with optional channel and hook filters. Use to select corpora; use search_clips for timed transcript windows, search_library for bounded revision-bound retrieval, or query_by_facets for exact stored labels. Returns ranked item snippets and search-mode metadata. Semantic modes depend on the local index and helper; unavailable semantic search falls back to keyword with a reason.

search_clipsA

Find transcript windows in saved items, with timing, source links and ranked snippets. Use to locate quotable passages; use search_uoinks for whole items or search_library for bounded clip-first results with resource URIs. Long source cues keep coarse timing and link to the cue start. Semantic modes can fall back to keyword; inspect the returned search-mode metadata and passage verification before quoting.

get_evidence_cardA

Build a saved item's evidence card with identity, summary hints and clips spread across its timeline. Use when you need a selectable card profile or clip count; use get_library_item for a bounded default Librarian card with revision-bound resource URIs. Supply at least one selector: a matching slug takes precedence, then video_id is tried. Missing items return an error; cards expose no local file paths.

get_uoink_corpusA

Read a saved item's complete Markdown corpus, citation rows and available source or image pointers by slug. Use when you need the full local document; use get_library_item then read_library_resource for bounded chunks, or get_citation_map for citations alone. Returns an error if the corpus is missing or unreadable. Image pointers refer to saved pixels; this call performs no image interpretation.

analyze_commentsA

Analyze a saved capture's comments with Anthropic and replace its stored comment-intelligence section. Use to refresh themes, mentioned products and disagreements; use get_playbook_input with comment-mining for local evidence that your own agent analyzes. Requires a saved Anthropic key and at least five saved comments. Sends comment content to the provider and returns the resulting analysis lists.

classify_hookA

Classify a saved capture's opening with Anthropic and replace its stored hook analysis. Use to run the classifier; use get_taxonomy to inspect saved classifications or classify_facets to store your own labels. Requires a saved Anthropic key and sends saved title, transcript and comment context to the provider. Returns the hook, explanation, confidence and correction count, and appends a taxonomy entry.

get_taxonomyA

Read saved hook-classification rows, optionally restricted to a channel or hook. Use to inspect prior classifier results; use get_facet_taxonomy for allowed labels, classify_hook to run the classifier, or analyze_self_channel for own-channel trends. Returns a bounded taxonomy list without rerunning analysis.

get_citation_mapA

Read all stored transcript and screenshot citations for one saved capture. Use for citation timing and source links without the full corpus; use get_uoink_corpus for Markdown or export_cited_range for a bounded cue-aligned export. Returns separate transcript and screenshot lists with source-aware deep links where available. Requires a saved video_id in the item's sidecar; missing captures return an error.

get_uoink_healthA

Read the stored per-section extraction health of a saved capture. Use to diagnose missing or failed capture sections; use get_transcript_reliability for questionable transcript spans. Reads the index snapshot, falling back to the sidecar, and returns an error when no health data exists. It does not retry extraction or recompute health.

find_mentionsA

Find recorded entity-name mentions across saved items, with stored spans, timing and source links. Use for people, products, tools or topics already indexed as entities; use search_clips for arbitrary transcript text or search_uoinks for whole-item search. Ranks by name match with newest first on ties. Results are limited and do not extract new entities.

get_transcript_reliabilityA

Read stored transcript-reliability spans for a saved item. Use to inspect suspect wording before quoting; use get_uoink_health for extraction-section health or export_cited_range for stored cue text. Returns status not_computed with empty spans when no reliability snapshot exists. Computation requires the helper endpoint or the user's auto-check setting; this read never downloads a model.

add_podcast_feedA

Register a podcast RSS feed for ongoing metadata monitoring. Use for the podcast feed and episode workflow; use register_source for the general standing-source interface or list_podcast_feeds to find an existing feed. Returns the existing row for a duplicate URL. Audio auto-ingest is a separate explicit opt-in, off by default; standing capture still requires source consent through the dashboard.

list_podcast_feedsA

List registered podcast feeds newest first, including their saved polling settings. Use to obtain a feed_id before polling or removal; use list_podcast_episodes to inspect episodes and list_sources for standing-source consent. Returns feeds and their count. This call reads saved registrations without fetching RSS.

remove_podcast_feedA

Stop monitoring a registered podcast feed. Use after list_podcast_feeds identifies the feed; use set_source_consent to disable capture while retaining detection. A source-linked feed is archived and disabled; a legacy unlinked feed is deleted with its tracked episode rows. Returns whether a row was affected; already published corpora are not removed.

poll_podcast_feedA

Refresh metadata for one registered podcast feed through the source-aware polling path. Use after add_podcast_feed when new episode observations are needed; use list_podcast_episodes to read cached episodes without a fetch. Managed sources honor detection due times and leases. Polling records observations; capture remains subject to separate source consent and scheduling. Returns the poll outcome or a feed error.

list_podcast_episodesA

List tracked podcast episodes newest published first, optionally scoped to a feed or lifecycle state. Use to obtain episode IDs and check download or transcription state; use list_podcast_feeds for feed settings or poll_podcast_feed to refresh observations. Returns episodes and count from local storage without fetching audio.

download_podcast_episodeA

Download a tracked podcast episode's audio to its canonical local MP3. Use after list_podcast_episodes, before transcribe_podcast_episode; use episode_to_corpus only after a transcript exists. Requires yt-dlp and ffmpeg, and blocks until download finishes or fails. Reuses an existing nonempty canonical file and records its path on the episode.

get_whisperx_statusA

Inspect local WhisperX availability, the selected model and the diarization default. Use before transcribe_podcast_episode to identify missing runtime prerequisites; use get_job_status to monitor an already queued transcription. Returns availability and supported model settings without downloading a model or starting transcription.

transcribe_podcast_episodeA

Queue local WhisperX transcription for a downloaded podcast episode. Use after download_podcast_episode and get_whisperx_status; monitor the returned job with get_job_status, then publish with episode_to_corpus. A single worker writes or replaces the JSON transcript beside the audio. First-time model downloads require user consent: only reissue with consent_given after the dashboard records the opt-in. Missing runtime returns an error.

episode_to_corpusA

Publish a completed podcast transcript as a searchable local corpus with source-aware citations. Use after transcribe_podcast_episode completes; use download_podcast_episode for audio acquisition. Writes deterministic Markdown and sidecar files, replaces the prior corpus/index snapshot and repairs partial previous attempts. Missing audio or transcript prerequisites return an error; success identifies the published item.

get_library_activityA

Report saved-library capture activity, shelf churn and standing-source observations for a time interval. Use for counts and revision-bound drill-down; use get_library_brief_input for a daily writing packet or source_status for one source's observations. Start with a summary, then reuse its report_revision for detail pages. Returns deterministic metrics and evidence; stale detail revisions require a fresh summary.

search_libraryA

Retrieve a bounded, clip-first selection of saved-library matches with revision-bound card and excerpt URIs. Use for citation-ready research; use search_uoinks for whole-item filters or search_clips for more transcript windows. Check passage_verified before quoting; citation_reason explains missing passages. Follow resource URIs with read_library_resource and next_step for more retrieval. Results are not exhaustive; search-mode metadata explains keyword fallback.

get_library_itemA

Resolve one saved item to its bounded default Librarian card and canonical card, excerpt and initial corpus-chunk URIs. Use before quoting or paging through evidence; use get_evidence_card for selectable profiles and clip counts, or get_uoink_corpus for the full local Markdown. Supply exactly one selector. Follow returned URIs with read_library_resource; missing or stale evidence produces a structured refusal.

read_library_resourceA

Read a bounded saved-library card, excerpt, corpus chunk, shelf page or brief from its uoink URI. Use as the tools-only equivalent of resources/read, after search_library or get_library_item supplies a URI. Use get_uoink_corpus for a full document instead of chunk paging. Validation, rendered contents and refusals match native resource reads; follow next_step on stale revisions or unavailable evidence.

get_playbook_inputA

Prepare cited local evidence and instructions for a client-written research report. Use for creator comparisons, transcript analysis, comment mining or repurposing; use analyze_self_channel for computed own-channel performance, search_library for ad hoc retrieval and get_library_brief_input for a dated operational brief. Returns instruction, packet and packet_hash within the library read budget. The client does the analysis; no model is called. Argument selectors depend on the selected playbook.

discover_candidatesA

Plan or run a ScrapeCreators search and save uncaptured candidates. Use to find new material; use search_library for saved evidence and queue_discovered to capture selected candidates. Off by default; requires a saved key, discovery enabled and daily credit room. Omit confirm_credits for a free plan, then confirm its exact cost. Spending requires agent discovery consent or a dashboard intent; stdio needs agent consent because dashboard intents are process-bound. No candidate is captured by this call.

queue_discoveredA

Enqueue selected saved discovery candidates for normal capture. Use after discover_candidates returns a run and candidate keys; use that tool again to search for new candidates. Requires agent discovery consent or a dashboard intent; stdio needs agent consent because dashboard intents are process-bound. Returns queued and skipped picks, skipping unknown, unsupported, captured or queued URLs. No ScrapeCreators credits are spent; the resident retry worker performs capture later.

get_library_brief_inputA

Prepare a dated library-status packet for a brief that the client writes. Use for a requested daily brief; use get_library_activity for direct metrics or get_playbook_input for topical research. Returns job_key, input_hash, bound revisions, counts, coverage, up to 20 work rows and five Librarian cards within 24,576 bytes. No work lease is acquired. Pass the unchanged packet to publish_library_brief after writing.

publish_library_briefA

Save a client-written daily brief as an immutable local artifact. Use only for the user's requested brief job after get_library_brief_input; use read_library_resource to read a published brief. Validates the exact packet against current data and binds quotes to supplied evidence. An identical submission_key retry returns the saved result; stale_brief requires fresh input. Writes under reach/briefs without applying labels or changing the assignment queue.

export_cited_rangeA

Export verbatim stored transcript cues with speaker provenance, overlapping chapters, safe links and evidence references. Use after search_clips identifies a passage; use get_citation_map for all citation rows or read_library_resource for an existing resource URI. Select exact cue-aligned bounds or a current excerpt, at most 120 seconds and 200 cues. Returns source/media revisions; refusals include next_step. It does not fetch media or recompute transcription.

get_transcriptA

Get the transcript of one YouTube video, with honest provenance. Use when you need what a video says; use search_clips to find passages across saved items or uoink_video to save a video with screenshots. Order: the saved library copy (no network), then captions fetched on this computer's own connection, then local Whisper on the audio when YouTube refuses captions (429, bot check) or has none. Returns source (library, captions, auto_captions or whisper), language, a plain why on any fallback, and the text; long transcripts page with next_cursor. Nothing is added to the library.

list_library_workA

List Librarian assignment counts and one cursor page of work for a run. Use to inspect readiness before claim_library_work; use get_library_activity for aggregate library metrics. Listing expires stale leases and can update work state. Returns run revision, work states, manifest dispositions and waiting-client status. The server does not perform the Librarian's model work.

claim_library_workA

Lease Librarian evidence packets, or renew, release or cancel your current attempts. Use after list_library_work to acquire work; use submit_library_result to store an assignment. Claim returns up to twelve packets with attempt tokens. Renew/release/cancel require the work_id and attempt_token from that claim. Leases last 60-900 seconds, never beyond one hour from first claim; three attempts exhaust a row.

submit_library_resultA

Store a client's shelf-assignment result for one currently leased Librarian item. Use after claim_library_work; use apply_reshelving to preview accepted assignments as label changes. Copy the packet's identifiers and revision hashes unchanged, and quote its evidence. Submission stages results without changing current labels; an identical submission_key retry replays the recorded response. Report unavailable usage honestly instead of inventing zero usage.

apply_reshelvingA

Create a reshelving preview or apply an approved preview to current labels. Use after submit_library_result stages assignments; use pin_shelf for a user's individual shelf decision or undo_library_apply to reverse the latest apply. Preview needs a run and current projection revision; apply needs its preview_id, delta_hash and operation_key. Apply is disabled by default and stops at 15 percent churn without trusted human approval; expired or stale previews must be rebuilt.

pin_shelfA

Record a user's pin, unpin or move decision for one item's shelf membership. Use for an explicit individual decision; use apply_reshelving for run-wide proposed changes. Requires a dashboard-minted user_intent_token tied to the exact operation and current projection revision; no actor string grants authority. The Librarian cannot override the resulting user decision. Retry the same operation_key only with identical arguments.

undo_library_applyA

Reverse the latest applicable library apply by replaying its stored inverse as a new journal operation. Use to undo the apply that produced the current projection revision; use apply_reshelving for a new proposal or pin_shelf for one item's shelf choice. Requires a dashboard user_intent_token. The inverse is itself reversible; identical operation_key retries replay the receipt, and stale revisions are refused.

list_sourcesA

List standing RSS, YouTube channel and playlist sources with consent, enrollment, daily allowance and detection health. Use to find source_id values and compare source states; use source_status for one source's observations or list_podcast_feeds/list_monitored_playlists for legacy feed and playlist IDs. Returns a keyset page from local storage. Pass its cursor unchanged to continue; no poll or capture starts.

register_sourceA

Register a standing source for future metadata detection with capture consent off. Use for the general source workflow; use add_podcast_feed or add_monitored_playlist when you need their feed or playlist IDs. Creates a due detection cursor without an immediate network call; the resident scheduler polls later. Duplicates return the existing source unchanged with created=false. Enable capture separately through set_source_consent after dashboard confirmation.

source_statusA

Read one standing source's summary, observed-item page and all in-flight reservations. Use after list_sources for detailed capture and classification state; use set_source_consent to change capture permission. Returns saved observations without enrolling, polling or starting work. Continue item pages with the returned cursor; revisions in the summary are needed for consent changes.

set_source_consentA

Enable or disable standing capture for a registered source after local confirmation. Use after register_source and source_status; use remove_podcast_feed or remove_monitored_playlist to stop their monitoring entirely. Requires current revisions and a dashboard-minted intent. Enabling permits at most 25 initial back-catalog items and 10 starts per UTC day; disabling releases unstarted reservations. Identical operation-key retries return the stored receipt; active captures are not cancelled by disabling.

analyze_self_channelA

Summarize the user's saved own-channel videos into hook and format history, monthly performance trends and top performers. Use for own-channel planning; use get_playbook_input with creator-teardown for a chosen creator's research packet or get_engagement_signal for one item's local usage. Scope by a registered handle or omit it for all recognized own channels. Results reflect saved data, not live channel analytics.

get_schema_versionA

Report the local SQL migration, item and sidecar schema versions and supported read ranges. Use before interpreting older saved records in an integration; use get_uoink_health to diagnose a capture's contents. Returns null for the SQL version if it cannot be read. This is a storage compatibility check, not a migration request.

get_user_roleA

Read the saved user role and its resolved dashboard sort and filter emphasis. Use to inspect the current dashboard preference; use set_user_role to change it or get_user_taste for writing and content preferences. Returns the normalized role, emphasis and supported roles without changing settings.

set_user_roleA

Replace the saved user role that controls dashboard sort and filter emphasis. Use for the user's chosen working role; use get_user_role to inspect it or update_user_taste for content preferences. Returns the normalized role and computed emphasis. Other saved settings are retained.

check_live_statusA

Fetch video metadata to determine whether a broadcast is live, upcoming or finished. Use before deciding to capture now or wait; use uoink_video or uoink_url to capture, and get_job_status for an existing job. Requires the metadata-fetch runtime and network access. Returns the canonical URL, title and detected live state without starting extraction.

add_monitored_playlistA

Register a YouTube playlist for ongoing metadata monitoring. Use for repeated observation; use uoink_playlist for a one-time capture batch or register_source for the general source interface. Returns the existing row for a duplicate URL. Managed playlists start with capture consent off and clamp polling to at least 15 minutes; capture requires separate source consent and the resident helper.

list_monitored_playlistsA

List registered playlist monitors newest first with saved monitoring settings. Use to obtain playlist_id before polling or removal; use list_monitored_playlist_events for discoveries and list_sources for standing-source consent. Returns playlists and count from local storage without fetching YouTube.

remove_monitored_playlistA

Stop monitoring a registered playlist. Use after list_monitored_playlists identifies it; use set_source_consent to stop capture while keeping metadata detection. Source-linked playlists are archived and disabled; legacy unlinked playlists and their discovery events are deleted. Returns whether a row was affected. Saved captures remain available.

poll_monitored_playlistA

Refresh observations for a registered playlist through its source-aware polling path. Use after list_monitored_playlists when fresh observations are needed; use list_monitored_playlist_events for cached discoveries. Stdio requires the resident Uoink helper. Managed sources honor due-time and lease gates, and detection does not enqueue captures; captures require separate source consent and reservations. Returns the poll outcome and observation summary.

list_monitored_playlist_eventsA

List recorded playlist-discovery events newest first, optionally filtered by playlist or lifecycle state. Use to inspect what monitoring found; use poll_monitored_playlist to refresh observations or list_monitored_playlists for monitor settings. Returns saved events and count. Reading events does not enqueue or retry captures.

get_user_tasteA

Read consolidated TASTE.md preferences and performance anchors. Use for writing or example-selection context; use get_user_memory for hand-edited personal notes or update_user_taste to replace a preference section. If missing, regenerates the file from local engagement and anchors, including a configured vault mirror. Returns content and path without calling a model.

get_user_memoryA

Read the user's hand-edited USER.md notes about channels, topics and workflow. Use for personal context; use get_user_taste for consolidated content preferences. Creates a local skeleton on first read when absent, then returns content and path. Existing hand-edited text is preserved; the taste consolidator does not overwrite it.

update_user_tasteA

Replace one persisted taste-preference section and regenerate TASTE.md. Use for a user-requested preference change; use get_user_taste to inspect current preferences or get_user_memory for separate personal notes. The supplied Markdown replaces that section verbatim. Regeneration updates the local file and configured vault mirror; other preference sections remain intact.

get_engagement_signalA

Read one saved item's time-decayed local engagement score, event counts and last-event time. Use to assess how the user has used an item; use analyze_self_channel for captured channel-performance trends or query_by_facets for stored performance labels. The score uses local interaction events with a 30-day half-life, not live social-network engagement.

classify_facetsA

Store your agent's facet classifications and tags on a saved item. Use after analyzing its evidence; use query_by_facets to retrieve labeled items or classify_hook for provider-run hook classification. Validates supplied labels, replaces those facet values and adds deduplicated tags. Missing performance tier and length bucket can be derived from saved metadata. Returns applied facets and tag counts; no model call runs here.

query_by_facetsA

List saved items matching all supplied facet and tag filters, newest first. Use for exact stored-label selection; use search_uoinks for text relevance, classify_facets to write labels, or get_facet_taxonomy to discover supported axes. Returns yoinks and count from local storage. Filters are AND-combined; omitted filters leave that axis unrestricted.

get_facet_taxonomyA

Return supported format, performance-tier, length-bucket and hook label lists. Use before classify_facets or query_by_facets; use get_taxonomy for saved per-item hook-classification rows. These are the server's allowed label constants, not counts of labels present in the library.

assemble_workspaceA

Select and rank saved examples for planning a video, with audience questions and available taste or own-channel context. Use to assemble a corpus slice; use get_workspace to read a saved workspace or critique_against_corpus to review a draft. Ranking uses format, performance tier and local engagement. Supplying workspace_id replaces that workspace's selected-item list; omitting it only returns the slice. Your agent performs any subsequent model analysis.

critique_against_corpusA

Prepare workspace grounding for a draft critique, or append your agent's findings to its critique log. Use after assemble_workspace; use get_workspace to read earlier critiques or revise_script to save a script revision. Omit findings to retrieve context, then submit your analysis with findings. Repeated submissions append log entries. Server-side BYO generation is not implemented; the agent performs the analysis.

list_workspacesA

List saved video-planning workspaces newest first. Use to obtain a workspace ID for get_workspace, assemble_workspace or generate_script; use list_scripts for script-version history. Returns workspace rows and count. This tool does not create workspaces or assemble new examples.

get_workspaceA

Read one saved video-planning workspace and its full critique log. Use after list_workspaces; use assemble_workspace to select new examples or critique_against_corpus to append findings. Supply id or its workspace_id alias; id takes precedence. Returns workspace and critique_log, or an error if the workspace is missing.

extract_claimsA

Store checkable claims that your agent extracted from a saved item. Use after reading its evidence; use verify_claim to record supporting or conflicting evidence and list_claims to inspect existing claims. Appends claim rows in extracted state and returns their IDs. The agent performs decomposition; this tool does not determine truth. BYO model execution is not implemented.

verify_claimA

Record your agent's evidence and alignment signals for an existing claim. Use after extract_claims; use get_claim to inspect evidence without changing it. Replaces the stored evidence and updates review status. Signals describe support, contradiction, mixed evidence or inconclusiveness, never a truth verdict; the user judges the claim. The agent performs research, and BYO model execution is not implemented.

list_claimsA

List saved checkable claims, optionally filtered by source item or review status. Use to select claim IDs for get_claim or verify_claim; use extract_claims to add claims. Returns claim rows and count. A verified status records that evidence was submitted, not that the claim is true.

get_claimA

Read one saved claim with its recorded evidence and alignment signals. Use after list_claims; use verify_claim to replace its evidence. Supply id or its claim_id alias; id takes precedence. Returns the claim or a not-found error. Review status and alignment signals are evidence records, not automatic truth verdicts.

generate_scriptA

Prepare workspace grounding for a video script, or save an agent-written script as a new version. Use for a new draft; use revise_script to revise an existing script or write_blog/write_tweet for prose pieces. Omit script to retrieve context, then submit your structured draft. The agent writes it; server-side BYO generation is not implemented. Returns grounding or the saved script ID and version.

revise_scriptA

Prepare revision context for an existing video script, or save its replacement as a linked new version. Use after get_script or critique_against_corpus; use generate_script for a new draft. Without revised_script, returns the previous script and grounding; with it, persists a new row whose parent is the prior script. The agent performs rewriting; server-side BYO generation is not implemented.

get_shot_listA

Derive and save a default shot list from a script's beats and its workspace format. Use when you want heuristic camera cues; use get_script to read the existing shot list without replacing it, or supply a custom shot_list through generate_script. This call overwrites the prior shot list and returns the generated rows. It does not run a model.

list_scriptsA

List saved video-script versions newest first, optionally within one workspace. Use to find a script ID for get_script or revise_script; use list_writing_pieces for tweet, thread and blog drafts. Returns script rows and count without generating or changing a draft.

get_scriptA

Read a saved video-script version, including beats, shot list and source-item citations. Use after list_scripts; use revise_script for a new version or get_shot_list to regenerate camera cues. Supply id or its script_id alias; id takes precedence. Returns the script object or a not-found error.

write_tweetA

Prepare grounding for a tweet or thread, or save a credited draft in the local Writing Studio. Use for short social drafts; use write_blog for articles or generate_script for video scripts. Without body, returns grounding unless direct generation is requested with the saved Anthropic key. With body, saves the agent's draft and returns soft Voice DNA warnings. Creator credit cannot be suppressed. Direct generation sends context to Anthropic and retries fatal phrase violations once before refusing to save; nothing is posted to a social account.

write_blogA

Prepare grounding for a blog article, or save a credited Markdown draft in the local Writing Studio. Use for long prose; use write_tweet for tweets and threads or generate_script for video scripts. Without body, returns grounding unless direct generation is requested with the saved Anthropic key. With body, saves the agent's draft and returns soft Voice DNA warnings. Keep the required Source credit. Direct generation sends context to Anthropic and retries fatal phrase violations once before refusing to save; nothing is published to a website.

list_writing_piecesA

List saved tweet, thread and blog drafts newest first, optionally filtered by kind or source item. Use to find a piece ID for get_writing_piece; use list_scripts for video-script versions. Returns pieces and count. The list contains local drafts, not posts fetched from social or publishing accounts.

get_writing_pieceA

Read one saved Writing Studio tweet, thread or blog draft. Use after list_writing_pieces; use write_tweet or write_blog to save another draft, or get_script for video scripts. Supply id or its piece_id alias; id takes precedence. Returns the piece object or a not-found error.

add_style_anchorA

Save a named writing-style reference from pasted text or a URL. Use to guide later write_tweet/write_blog drafts; use list_style_anchors to select existing references. URL references fetch prose through the page extractor and can retain null raw_text if extraction fails. Returns the anchor; more than ten anchors is refused. Style references do not authorize copying source facts or suppressing creator credit.

list_style_anchorsA

List saved writing-style references with active flags and the configured ten-anchor cap. Use to select style_anchor_ids for write_tweet or write_blog; use add_style_anchor to save a reference or remove_style_anchor to delete one. Returns anchors and count without refetching URL contents.

remove_style_anchorA

Delete a saved writing-style reference. Use after list_style_anchors identifies the reference; use add_style_anchor to save a replacement. Supply anchor_id or its id alias; anchor_id takes precedence. Returns the selected ID and whether it was removed. Previously saved writing pieces remain.

uoink_urlA

Capture a supported video URL through platform routing, with metadata/transcript fallback for other yt-dlp sites. Use for non-YouTube videos or long-video modes; use uoink_video for a single standard YouTube capture, uoink_page for webpage prose, uoink_x for X text chains, or uoink_reddit_thread for discussion comments. Requires the local media runtime. Saves a local corpus; generic fallback has no screenshots, comment fetch or hook classification. Recapturing a source can replace its saved corpus and metadata.

uoink_noteA

Save user-provided text as a local searchable note with title and attribution. Use for pasted ideas or prose without fetching a source; use uoink_page for a web article or uoink_image for pixels. Returns the saved item ID, slug and metadata. Empty or invalid notes are refused; title can be derived from the text.

uoink_imageA

Save supplied PNG, JPEG or WebP bytes as a local image corpus item with optional caption and attribution. Use for screenshots, diagrams or images; use uoink_note for text or uoink_page to fetch a webpage. Accepts strict base64 or a data URL and returns the saved item ID, slug and metadata. Invalid images are refused. No cloud OCR or vision service runs.

uoink_xA

Capture an X post and its author's earlier chain as a local text corpus item. Use for post text; use uoink_url for an X video capture or uoink_page for general webpage prose. Requires x_text_capture_enabled and public source access. Returns the saved item ID, title, captured-tweet count and metadata; disabled capture or extraction failure returns an error. Recapturing a source can replace its saved corpus and metadata.

uoink_pageA

Fetch an allowed webpage as Markdown and save it as a local corpus item. Use for articles and documentation; use uoink_url for videos, uoink_x for post chains or uoink_reddit_thread for comments. Add an unlisted host with add_allowed_site only with user authorization. Crawl4AI renders locally when available; static fallback has no screenshot. Returns extracted content and metadata; video_id is null if extraction succeeded but persistence failed. Recapturing a source can replace its saved corpus and metadata.

uoink_reddit_threadA

Capture a Reddit post and filtered comment tree as a local discussion corpus. Use when replies matter; use uoink_page for general webpage prose or uoink_url for video extraction. Fetches Reddit's public JSON without API credentials or the page allowlist, then saves nested replies with source_type reddit_thread. Returns captured content, comment counts and the saved item ID; fetch or save failures return an error. Recapturing a source can replace its saved corpus and metadata.

import_listA

Preview or import a list in batches of at most 50, skipping saved items and adding a source tag. Supply a HackerTube, HN thread, Reddit thread or YouTube playlist url; a local .csv/.txt file of YouTube Watch Later or Reddit saves; or items with source watch_later, x_bookmarks or reddit_saved. X status links also work in a text file. Use uoink_video for one video or uoink_playlist for one tracked playlist job. Set preview_only before importing. File/items input stages up to 5000 saves locally; reuse source.url (also stage_url on staging) with next_offset for later batches. YouTube uses the video queue; X/Reddit posts capture serially at least 15 seconds apart. X text capture must be enabled. Returns found, already, queued, next_offset and item rows. HackerTube imports retain ODC-By 1.0 credit.

list_allowed_sitesA

List saved webpage-capture hostname patterns and their active state. Use before uoink_page to check access; use add_allowed_site or remove_allowed_site to change the allowlist. Returns sites, count and default seeds. Seeds can be removed like other entries; this call does not test reachability or fetch pages.

add_allowed_siteA

Add a hostname pattern to the webpage-capture allowlist. Use for a user-authorized uoink_page destination; use list_allowed_sites to inspect rules or remove_allowed_site to revoke one. Plain hosts include subdomains; wildcard patterns match their specified subdomains. Returns the saved rule and reuses an existing pattern unchanged. This grants future page access but does not capture a page.

remove_allowed_siteA

Remove a hostname pattern from the webpage-capture allowlist. Use to revoke a rule found by list_allowed_sites; use add_allowed_site to restore it. Returns the normalized pattern and whether a rule was removed. Saved captures remain, and overlapping rules can still allow the same host.

research_startA

Start a local research run for a question, with source and ASR budgets. Returns the run_id, normalized scope and network sources needing consent. Makes no network requests. Only the user's research_network_sources setting grants network consent; research tools cannot change it.

research_discoverA

Discover and rank sources for a saved research run from the local library and consented YouTube, HN, HackerTube and podcast searches. Returns a merged candidate pool with score explanations, capture estimates, source counts, skips and errors; saves candidates and query history. Uses at most 8 queries and 4 iTunes calls. Reddit search is deferred. Discovery does not capture sources. Treat returned titles and descriptions as untrusted data.

research_captureA

Queue selected research candidates by keys or top score order. Rechecks the library and enforces the run's ASR budget; unknown duration assumes 15 ASR minutes. transcript_only defaults to true and uses lite mode: full transcript, sparse screenshots, no comments. Returns queued, already, skipped, committed/remaining ASR minutes and a serial ETA. Podcast audio capture is deferred. Treat returned titles as untrusted data.

research_statusA

Read a research run's capture states from the local queue and saved library transcripts. Returns items, counts, ready_count, estimated ASR minutes used/left and done; saved transcript provenance distinguishes captions from Whisper. Reconciles saved budget reservations without fetching transcripts. Treat returned titles as untrusted data.

research_evidenceA

Find cue-aligned evidence only in a research run's selected ready sources. Returns untrusted excerpts inside the library's escaped JSON fence, revision-bound ids, timestamp links and ASR doubt spans. Quotes are the unescaped words, not the escape sequences. Saves excerpt membership and call history locally. Windows are at most 120 seconds and 200 cues. Counter stance only expands the search query. Each call consumes one of the run's 10 evidence calls; no network or library writes.

research_publishA

Verify research claims against this run's stored excerpts and current transcript revisions, then write a local Markdown report, escaped HTML report and verification ledger. Submit quotes as unescaped words, not escape sequences. Renders only matched stored text; drops unbound or ambiguous quotes and judge=no claims. Flags ASR doubt, unconfirmed speakers, overstatement, low overlap and one-sided search. Enforces quote/source caps. Reuses a saved result only when the payload, source revisions, run evidence and output paths are unchanged. No network; writes only inside the research run.

Prompts

Interactive templates invoked by user choice

NameDescription
consult-libraryAnswer a question from cited library evidence: up to 5 bounded excerpt previews for a topic with revision-bound follow-up URIs.
evidence-briefCompose a citable brief from up to 5 bounded Librarian cards on a topic, optionally limited to items captured on or after a UTC date.
whats-newDeterministic counts and up to 20 recorded events (captures, shelf revisions, applied membership changes) for the last N days.
reshelve-reviewRead an existing stored reshelving preview for review: run, revisions, delta hash, expiry, changes, pins, exclusions and churn. Never applies or approves anything.
outlier-postsFind posts above a creator's local median, with cited metrics and hooks.
transcript-intelExtract hooks, ideas, quotes and CTAs from saved transcripts.
comment-miningMine saved comments for customer language or demand signals.
creator-teardownAnalyze a creator's saved posts, positioning, formats and cadence.
repurpose-packPrepare cited content atoms and credited drafts from saved captures.
thesis-checkTest a thesis against your saved sources: the assumptions they share, evidence for and against, and what's missing.
deep-researchResearch a question from videos and podcasts, then publish locally verified timestamped receipts.

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

A3.8/5.0

Scored across 99 tools

Disambiguation3/5

The descriptions frequently cross-reference alternatives, and the source-specific capture tools are well separated. Nevertheless, several crowded clusters retain overlapping purposes: search_uoinks/search_clips/search_library, the evidence-card/corpus/resource readers, general standing sources versus specialized feed/playlist monitors, and general discovery versus research discovery. The distinctions are real but require careful reading.

Naming Consistency4/5

All names use snake_case, and most follow readable verb_resource or resource_action patterns with consistent get_, list_, add_, and remove_ families. The uoink_* brand-as-verb captures, noun-first research_* names, and outliers such as episode_to_corpus and source_status prevent full uniformity but remain predictable.

Tool Count1/5

At 99 tools, the surface is far beyond a well-scoped MCP set and spans several products' worth of capture, monitoring, research, library-management, and writing workflows. Many tools are individually useful, but the overall count creates an extreme discovery and selection burden.

Completeness3/5

Many major pipelines are covered end to end, including capture, citation, podcast transcription, research, reshelving, and drafting. However, there is no apparent workspace-creation operation despite the workspace-dependent planning flow, and saved items and writing drafts lack general deletion or fuller lifecycle management; these are notable gaps rather than a severely incomplete surface.

Maintenance

ActivityActive
ResponsivenessNo recent community issues