MCP App Store
by KeenEthics
ThoughtLeaders - YouTube Data icon

ThoughtLeaders - YouTube Data MCP Server for Claude and ChatGPT

By ThoughtLeaders

Overview

ThoughtLeaders turns YouTube into a database Claude can query: millions of tracked channels, hundreds of millions of indexed videos, and full transcripts you can search word for word, growing every day. Many uses; the one we have perfected is sponsorship intelligence. WHAT IS IN THE DATA Transcripts. Full text, timestamped to the second, searchable across the corpus. Performance over time. Daily snapshots of views, likes and comments for videos, and views and subscribers for channels. Brand mentions. Detected mentions tagged organic or sponsored, with snippet and timestamp. Channel intelligence. Upload history, format mix across long-form, Shorts and live, category and language, plus enriched signals per channel: projected views at day 30, renewal rate (how often sponsors come back), fulfillment rate, sponsorship score, and evergreenness. Similarity and discovery. Look-alike channels, channels that fit a brand's audience, and top channels for any topic or trait. BUILT FOR SPONSORSHIP INTELLIGENCE Brands buying YouTube sponsorships: vet a creator's track record, see which competitors already sponsor them, and find look-alike channels before you spend. Creators and talent agents selling: benchmark against comparable channels, show brands the renewal and fulfillment record, and find the brands already active in your niche. YOUR OWN WORKSPACE If your team runs sponsorships on ThoughtLeaders, the same connection reads your organization's own campaigns, reports, proposals and sponsorship records. WORKS BEST WITH THE SKILLS The free, open-source ThoughtLeaders CLI ships skills that teach Claude the schemas, query patterns and vocabulary behind these tools. Install with pipx install thoughtleaders-cli, then tl setup claude. Repo: github.com/ThoughtLeaders-io/thoughtleaders-cli. GETTING STARTED Sign up free at thoughtleaders.ai and connect. Queries draw on a monthly credit balance; the free tier includes a starter allowance.

How to connect

  1. Claude: Open the ThoughtLeaders - YouTube Data listing in the Claude connectors directory and choose Connect. Claude listing ThoughtLeaders - YouTube Data exposes 26 tools here.
  2. ChatGPT: Open the ThoughtLeaders - YouTube Data listing in the ChatGPT app directory and choose Connect. ChatGPT listing ThoughtLeaders - YouTube Data exposes 27 tools here.

Tools

tl_agent_complete

ChatGPT
Call once at the end of a ThoughtLeaders skill (a /tl-... slash command) that a ThoughtLeaders surface handed to this agent. Pass the skill's slug and how the run ended: 'completed' (the deliverable was produced), 'no_results' (it ran but had nothing to show) or 'failed'. Free; does not cost credits. A skill started by hand has no run to close and gets a not-found answer, which is fine to ignore.

tl_balance

ChatGPT
Returns the organization's current credit balance and the ten most recent usage entries (endpoint, results count, credits charged). Call this before any expensive query to make sure the org has enough credits to cover the cost — every data query consumes credits, and a 402 response means the balance was insufficient. Free; does not cost credits. Recommended as a sanity check after tl_whoami and again after any unusually large query.

tl_brands_find

ChatGPT
Resolve an arbitrary brand reference (name, slug, website domain, full URL, internal numeric ID, or keyword) to a single canonical brand record with its integer ID. The lookup searches across name, slug, website (domain extracted), and the brand's keyword fields. When the input is ambiguous, the response carries candidates: [{id, name, website}] so the caller can re-resolve with a tighter query or pick an ID directly. When no candidates match, the response is a not-found error. Use the returned id as the brand identifier in tl_db_es queries (e.g. sponsored_brand_mentions: <id>) or in tl_db_pg joins. Costs the brand-detail credit rate (see tl_describe brands).

tl_brands_similar

ChatGPT
Find brands similar to a given brand via vector KNN.Use to expand a set of comparable advertisers. brand is any brand reference (name, slug, website, or ID). Optional: limit (1-100, default 20) and mbn ('yes'/'no'/'both', default 'both') to restrict to Media-Buying-Network brands.

tl_channels_find

ChatGPT
Resolve an arbitrary channel reference (partial name, slug, YouTube channel URL, video URL, @handle, external channel ID, or internal numeric ID) to a single canonical channel record with its integer ID. When the input is ambiguous, the response carries candidates: [{id, name, subscribers}]. A channel that is not in the ThoughtLeaders index returns a not-found error; this tool does not add channels to the index. Use the returned id in tl_db_es queries (filter on channel.id) or in tl_db_pg joins. Costs the channel-detail credit rate (see tl_describe channels).

tl_channels_lookalike

ChatGPT
Find audience/topic look-alikes for a seed channel using BrandChannelMatcher's methodology: a vector search over the channel's audience embedding (MSN channels only) and topic-description embedding, re-weighted by ~15 channel-similarity factors (keyword overlap, demographics, posting cadence, average video duration, political/gender alignment, …). This is the exact 'lookalike' primitive the brand→channel matching methodology uses — prefer it over tl_channels_similar (plain feature-vector KNN) when the goal is sponsorable look-alikes for a campaign. channel is any channel reference. Optional: limit (1-100, default 20), msn/tpp tri-state ('yes'/'no'/'both', default 'both'), created_since (only channels created on/after a YYYY / YYYY-MM / YYYY-MM-DD date). Each row carries the 0-100 methodology score, es_score, sponsorship_score, has_demographics, and msn/tpp flags.

tl_channels_similar

ChatGPT
Find channels whose ~200-dim feature vector is closest to a seed channel's (generic Elasticsearch KNN over category + demographic signals). Use this for broad 'more channels like this one' discovery. For sponsorable audience/topic look-alikes scored by the brand→channel methodology, use tl_channels_lookalike instead. channel is any channel reference (name, @handle, URL, external or internal ID). Results always share the seed channel's format. Optional: limit (1-100, default 20), language (a single code, or '*' for every language; left unset it follows the seed channel's own language, with 'en' also admitting channels of undetected language), min_score (0-1 similarity cutoff; defaults to the cutoff the ThoughtLeaders channel page and Chrome extension use), msn/tpp tri-state filters ('yes'/'no'/'both', default 'both'). Optional exclude: comma-separated channel ids already seen — returns the next page of results and bills only those new rows (use for load-more). Rows are enriched with subscribers, projected views, category, a representative CPM, and an audience summary.

tl_credits_history

ChatGPT
Returns the paginated history of credit top-up purchases for the caller's organization (purchase date, USD amount, credits granted, status). Use this to verify a payment landed or to surface receipts. limit defaults to 25 (max 100); offset paginates older purchases. Free; does not cost credits.

tl_db_es

ChatGPT
Execute a read-only Elasticsearch search over the main article / channel index. query is the full ES DSL request body written as a JSON string, such as {"size": 0, "query": {"match_all": {}}}. Accepted top-level keys include query, size, from, sort, _source, aggs/aggregations, track_total_hits, highlight. For bounded transcript excerpts, set the tool argument include_highlight=true and add highlight: {fields: {transcript: {fragment_size: 200, number_of_fragments: 3}}} to the query body. Use a small size and limit _source to identifying fields such as id and url. Returned excerpts are in each result's highlight.transcript; highlights are omitted by default and do not guarantee timestamps. Account access and credit charges still apply. Hard limits: size ≤ 500, from + size ≤ 10_000, body depth ≤ 16, total nodes ≤ 1000, at most one aggregation. BLOCKED query types: query_string, regexp, wildcard, fuzzy, more_like_this, has_child, has_parent, parent_id, scripts. Rewrite blocked queries using term / terms / match / bool / nested and multiple calls instead. Use search_after for deep pagination (no scroll / pit / knn). Set pricing=true for a flat-rate dry-run estimate (no execution); aggregations price on the agg-doc-cap, plain searches on size. Cost follows the curve 1 + rate * 0.126 * n^1.2. Call tl_schema_es first for the document shape reference.

tl_db_fb

ChatGPT
Execute a read-only Firebolt SELECT for time-series metrics. ACCEPTED QUERIES ARE NARROW: single table from {article_metrics, channel_metrics}; no JOIN / CTE / subquery / set-op / lateral; WHERE / HAVING only on indexed columns; the leading column of the table's primary index MUST be filtered with equality or IN over literals. Trivial-aggregation exception: all-aggregate projections without GROUP BY / HAVING may omit WHERE. LIMIT/OFFSET are optional. Anything outside this grammar returns 400 with a reason code. Call tl_schema_fb first to discover columns and the primary-index tuple per table. Set pricing=true for a flat-rate dry-run estimate based on the query's LIMIT ceiling (no execution). Cost follows the curve 1 + rate * 0.126 * n^1.2 over rows returned.

tl_db_pg

ChatGPT
Execute a read-only PostgreSQL SELECT against the caller's role-scoped sandbox schema. Accepts: a single SELECT statement, optional LIMIT and OFFSET (server defaults to LIMIT 50 OFFSET 0 when omitted; explicit LIMIT must be ≤ 500; explicit OFFSET ≥ 10_000 returns 403 OFFSET_TOO_DEEP); query length ≤ 50_000 chars; AST depth ≤ 64 nodes ≤ 5000; no ::reg* catalog casts; function calls limited to an explicit allow-list. Call tl_schema_pg first to discover the tables/columns the caller can see. Set pricing=true for a dry-run: the server runs EXPLAIN, returns a cost estimate (multiplier, per-row extra, expensive tables/columns, planner-estimated rows, upper-bound cost at the effective LIMIT), and DOES NOT execute the SELECT — useful before committing credits on a heavy query. Cost follows the curve 1 + rate * 0.126 * n^1.2, multiplied by per-table and per-column extras for expensive items (see tl_describe db).

tl_describe

ChatGPT
Returns the catalogue of CLI resources: which entities are queryable (sponsorships, channels, brands, uploads, reports, snapshots, recommender…), the fields each one exposes, the filter vocabulary each supports, and the credit rate / cost model per resource (flat-per-call, linear-per-result, or non-linear 1 + rate * 0.126 * n^1.2 curve). Pass resource to drill into one entity (e.g. resource='sponsorships' or resource='db'); omit it for the full index. Use this BEFORE composing a query so you pick the right resource and know whether the cost scales linearly or super-linearly with result count. Free; does not cost credits.

tl_pricing

ChatGPT
Returns USD per credit and currency for estimating a workload's cost. Public endpoint — does not require authentication or cost credits. Use tl_describe for per-endpoint credit rates and a query's pricing=True mode for a cost estimate.

tl_recommender_channels_for_brand

ChatGPT
Recommend channels a BRAND should sponsor: resolves the brand to its most-recently-vectorised profile, then runs a KNN of that profile's ideal vector against the channel index. The brand-scoped entry point to the recommender. brand is any brand reference. Optional: limit (1-100, default 20), language (default 'en'; '' for all), msn (True = MSN channels only, default False).

tl_recommender_inspect_brand

ChatGPT
Return a brand profile's ideal-vector breakdown — the per-tag target values the recommender matches channels against. Use to explain a brand's audience targeting or why certain channels rank for it or its vector. brand is any brand reference (resolved to its most relevant profile).

tl_recommender_inspect_channel

ChatGPT
Return a single channel's feature-vector breakdown — the per-tag values that drive every similarity / top ranking. Use to explain WHY a channel is (or isn't) surfaced by the recommender or to explain its vector. channel is any channel reference.

tl_recommender_similar_brands_to_channel

ChatGPT
The inverse direction: given a CHANNEL, rank the brands most likely to sponsor it by running the channel's feature vector against the profile-vector index and deduping by brand. Use to answer 'which advertisers fit this creator?'. channel is any channel reference. Optional: limit (1-100, default 20) and mbn ('yes'/'no'/'both'/'all', default 'all' = no filter) to restrict to Media-Buying-Network brands.

tl_recommender_similar_to_profile

ChatGPT
Recommend channels matching a specific PROFILE by its ideal vector — the profile-ID-scoped version of tl_recommender_channels_for_brand (use this when you already have a profile_id rather than a brand). profile_id is the integer profile ID. Optional: limit (1-100, default 20), language (default 'en'; '' for all), msn (True = MSN channels only, default False).

tl_recommender_tags

ChatGPT
List or search the vector-tag catalogue — the named dimensions of the recommender's feature space (content categories like 'Cooking', demographics like 'Age 18-24', geo shares like 'USA share', formats, etc.). Optional q filters by case-insensitive substring. ALWAYS call this first to get the exact tag name before tl_recommender_top_channels or tl_recommender_top_brands — those reject unknown tag names.

tl_recommender_top_brands

ChatGPT
Rank the top brands by a single vector tag, deduped to the highest-scoring profile per brand (one row per brand). tag MUST be an exact name from tl_recommender_tags. Optional: limit (1-100, default 50), msn/mbn tri-state ('yes'/'no'/'both', default 'both').

tl_recommender_top_channels

ChatGPT
Rank the top channels by a single vector tag (e.g. the channels most strongly 'Cooking' or skewed to 'Age 18-24'). tag MUST be an exact vector-element name from tl_recommender_tags. Optional: limit (1-100, default 50), msn/mbn tri-state filters ('yes'/'no'/'both', default 'both'). Returns rows with the per-channel tag value, plus the resolved tag / tag_index.

tl_reports_save_list

ChatGPT
Save a fixed list of entities as a new ThoughtLeaders report and get its link. entity sets the report type and what ids hold: channels (integer channel IDs, a channels report), brands (integer brand IDs), sponsorships (integer sponsorship IDs) or articles / uploads (video IDs in the composite <channel_id>:<youtube_id> form, the ES _id; a content report). The report shows exactly these IDs and never re-runs a search. title is at most 60 characters; description is one to three sentences saying what the list is. Resolve names to IDs first with tl_channels_find / tl_brands_find. Every call creates a new report, so call it once per list. Free.

tl_schema_es

ChatGPT
Returns the Elasticsearch document shape reference used by tl_db_es — a single canonical example article document plus notes on accepted top-level keys, size ≤ 500, from+size ≤ 10_000, body depth ≤ 16, total nodes ≤ 1000, blocked query types (query_string, regexp, wildcard, fuzzy, more_like_this, has_child, has_parent, parent_id), no scripts, at most one aggregation total. ES has no table concept, so there is no table parameter. Free; does not cost credits.

tl_schema_fb

ChatGPT
Returns the Firebolt schema for the two analytics tables exposed via tl_db_fb (article_metrics, channel_metrics), with each table's column types and its __primary_index tuple. The Firebolt backend is extremely strict: SELECT-only on these two tables (no JOIN, CTE, subquery, set-op, lateral) and the WHERE clause MUST filter on the leading column of __primary_index using equality or IN with literals. Pass table to fetch one table's row. Free; does not cost credits.

tl_schema_pg

ChatGPT
Returns the live PostgreSQL schema visible to the caller's role, rendered as YAML flow-mappings (one table per line). Each row lists the table name, every column with its type, and any COMMENT metadata. Roles: superusers see the full public schema; advertisers see sandbox_advertisers; publishers see sandbox_publishers. Pass table to fetch a single table's row instead of the whole listing. Use this BEFORE composing a tl_db_pg query so you only reference columns / tables that exist and are visible to the caller. The Postgres backend rejects queries that touch unknown tables, use blocked function calls, or reference columns that aren't in the caller's sandbox. Free; does not cost credits.

tl_usage_history

ChatGPT
Returns the paginated usage log for the caller's organization: every billable endpoint hit with timestamp, endpoint name, results count, credits charged, and balance-after. Use this to investigate cost spikes, audit which tools have been used recently, or reconcile the credit balance against expected charges. limit defaults to 50 (max 200); offset paginates older entries. Free; does not cost credits.

tl_whoami

ChatGPT
Returns the identity of the authenticated MCP caller. Use this as the first call in any session to confirm credentials are valid, learn which organization / plan / persona the caller belongs to, see their feature flags (intelligence, advertiser, publisher, full_access), and discover their organization's credit balance. The response also lists associated profiles in the same org and (for buyer-side users) the brands the caller can manage. Free; does not cost credits. Always safe to call. If this returns an authentication error, the human operator needs to complete the MCP OAuth login flow before any other tool will work.

tl_balance

Claude

tl_brands_find

Claude

tl_brands_similar

Claude

tl_channels_find

Claude

tl_channels_lookalike

Claude

tl_channels_similar

Claude

tl_credits_buy

Claude

tl_credits_history

Claude

tl_db_es

Claude

tl_db_fb

Claude

tl_db_pg

Claude

tl_describe

Claude

tl_pricing

Claude

tl_recommender_channels_for_brand

Claude

tl_recommender_inspect_brand

Claude

tl_recommender_inspect_channel

Claude

tl_recommender_similar_brands_to_channel

Claude

tl_recommender_similar_to_profile

Claude

tl_recommender_tags

Claude

tl_recommender_top_brands

Claude

tl_recommender_top_channels

Claude

tl_schema_es

Claude

tl_schema_fb

Claude

tl_schema_pg

Claude

tl_usage_history

Claude

tl_whoami

Claude

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/thoughtleaders"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/thoughtleaders)