AI tools & actions | Liquid Gold
Every tool and action your AI can use through Liquid Gold's MCP server, listed live: what it can do, what it can't, and the exact shape of each action.
Tools:
SEARCH_TOOLS — Find actions for a task and get a recommended plan.
GET_TOOL_SCHEMAS — Exact input and output schemas for actions SEARCH_TOOLS returned.
MULTI_EXECUTE_TOOL — Run up to 5 independent actions in one call; each result carries a log_id.
SUBMIT_FEEDBACK — Report an empty, wrong or failed result with its log_id, or bad content by category.
Actions:
CHANNEL_FIND — Find a published YouTube channel by creator name, handle, slug, or URL.
CHANNEL_OVERVIEW — Orient within one published channel: topic map, counts, last publication, and coverage.
CHANNEL_SEARCH — Search one channel's transcripts for what it actually says: quoted passages with timestamped source links.
PASSAGE_MATCH — Find passages by exact text: exact words, word stems or prefixes, a case-insensitive regex like `(?i)concurren\w*` or `cold (email|call)`, or a glob like `*cache*` (syntax: "glob"). Searches every published channel, or one with `channel`. The pattern needs at least 3 letters or digits in a row. Returns up to `limit` (default 10, max 50) timestamped quotes, each with its video and sourceUrl; never a count or total. Use CHANNEL_SEARCH for meaning-based search.
CATALOG_SEARCH — Find channels, topics and videos about a subject across Liquid Gold. Answers questions like "which channels best explain onboarding".
CHANNEL_TOPICS — A channel's topics with summaries, paged.
VIDEO_GET — Read one video's page: title, chapters, duration, publish date and its YouTube link.
VIDEO_PASSAGES — A video's transcript in reading order, paged; each passage carries a timestamped source link.
READ_CREDIT_USAGE — Read my own remaining credits, bounded usage analytics (the newest 1000 consume rows within 30 days; truncated=true means the totals cover only those observed rows, not the full window) and the newest 20 ledger rows. Free even at zero balance. Credits prevent abuse, not a paywall; email hello@bourdonlabs.com for more.
LIBRARY_LIST_SAVED — The signed-in member's saved channels, newest first.
LIBRARY_SAVE_CHANNEL — Save a channel to the member's library. Saving twice changes nothing.
RESEARCH_ASK — Ask a research question over the Liquid Gold corpus. The question check runs in the background (4 to 10 s); this action waits up to 10 s for it, then returns the request: up to 3 clarifying questions (state needs_clarification), a brief to approve (state awaiting_approval), or state checking when the check is slower (poll RESEARCH_STATUS). Nothing expensive runs until RESEARCH_APPROVE. Idempotent on idempotency_key.
RESEARCH_STATUS — Read a research request: state, brief, open clarifying questions, progress and outcome. Poll while state is checking, queued or running, and wait retryAfterSeconds between polls.
RESEARCH_ANSWER — Answer the open clarifying questions of a request in state needs_clarification: one answer per open question id. The re-check runs in the background; this action waits up to 10 s for it and returns the new state (more questions, or awaiting_approval), or state checking when it is slower (poll RESEARCH_STATUS).
RESEARCH_APPROVE — Approve the brief of a request in state awaiting_approval and queue the research run. Quote brief.version exactly; a stale version is refused. Counts against the member's daily research allowance.
RESEARCH_REPORT — Read the finished report of a research request (state complete): answer, findings with numbered citations, uncertainty, coverage and verified sources, as JSON fields plus the Markdown. The same report is the resource liquidgold://research/{id}/report.
RESEARCH_CANCEL — Cancel a research request that is not finished. Idempotent: a finished request answers with its state unchanged. A run that never started returns its daily allowance.
RESEARCH — Run a cited research report over the Liquid Gold corpus in one call. Mandatory fields: objective (the decision or outcome the research serves), context (who is asking, background, what they already know) and questions (1 to 5 concrete questions the report must answer). Optional: scope {channels, from, to, topics}, requirements (format, length, citations, include or exclude), depth (standard or deep), mode (sync waits up to about 45 s for the report, async returns once the run started) and idempotency_key. The brief is approved by this call and spends one research run of the member's daily allowance; a request that needs clarification returns its questions instead. Example: {"objective":"Choose a pricing model for our new design tool.","context":"We are a three-person startup launching a design plugin for small teams in 2027. Seat pricing is the norm we know; we have not decided between seats and usage-based pricing.","questions":["Which design-tool creators recommend usage-based over seat pricing, and why?","What problems do creators report after switching pricing models?"],"scope":{"from":"2024-01-01","topics":["pricing","SaaS"]},"requirements":["A comparison table of the arguments for each model","Quote at least two creators for each claim"],"mode":"async"}