TubeScout MCP¶

Local stdio MCP server (TypeScript, MIT) that turns YouTube into a research engine for AI agents - no API key required. Six tools: filtered video search, video metadata with engagement resonance, resilient transcripts (single or batch of 10), channel scans, and YouTube autocomplete as live search-demand data. Bundles six context-aware research skills for Claude Code, Codex and OpenCode: skeptic's video breakdowns, idea mining, niche validation, channel intel, tutorial-to-playbook, and demand-gap analysis. Install with npx -y tubescout, or as a Claude Code plugin via /plugin marketplace add not0lucky/tubescout. npm package published Aug 27, 2026 (v0.1.1).

Server type: Local (stdio, run via npx)
Auth: None (keyless - talks to YouTube's internal InnerTube API)
Install: npx -y tubescout (npm v0.1.1), or Claude Code plugin not0lucky/tubescout
Tools: 6 (all read-only)
Skills: 6 context-aware research methods (SKILL.md pack)
Repo: github.com/not0lucky/tubescout (MIT, created Aug 27, 2026)
Category: Content & Research
Built by: Anir (agramprojects.com)

Why This Matters for Operators¶

YouTube is where founders and builders show receipts - revenue dashboards, playbooks, real numbers on camera - but nothing mines it systematically. TubeScout makes YouTube queryable like a database: search with upload-window and duration filters, pull engagement signals (including likesPer1kViews resonance), and read transcripts at scale through a three-strategy fallback chain. The differentiating surface is demand data: get_search_suggestions exposes YouTube autocomplete as real keyword demand, and the bundled skills turn that into operator work products - niche validation, idea mining, demand-vs-supply gap analysis, and competitive channel intel. For market researchers, product operators and content strategists this is research tooling, not video plumbing.

Tools & Capabilities¶

Tool	What it does
`search_videos`	Search with filters: upload window, duration, sort by views or date
`get_video`	Full metadata plus engagement (`likesPer1kViews` resonance signal)
`get_transcript`	Plain-text transcript via a resilient 3-strategy fallback chain
`get_transcripts`	Batch transcripts (up to 10 videos), per-video error tolerant
`get_channel_videos`	Channel positioning plus recent uploads with view counts
`get_search_suggestions`	YouTube autocomplete = real search demand for keyword research

Skill pack (the research methods): /yt-breakdown (extract and stress-test every claim and number in a video), /yt-idea-mine (product ideas backed by demand signals and pains builders describe on camera), /yt-validate (go/no-go verdict: demand, saturation, competitor numbers), /yt-channel-intel (read a channel's strategy: cadence, outliers, what performs), /yt-playbook (turn a tutorial into executable steps adapted to your stack), /yt-gap (heavily searched topics served by weak or old videos - content plans and product angles). All skills are context-aware: they read the conversation for what you are building and tailor verdicts accordingly.

Installation¶

# Claude Code
claude mcp add --scope user tubescout -- npx -y tubescout

# Codex
codex mcp add tubescout -- npx -y tubescout

OpenCode: add to ~/.config/opencode/opencode.json under "mcp": "tubescout": { "type": "local", "command": ["npx", "-y", "tubescout"], "enabled": true }.

The skill pack installs separately: clone the repo and run ./scripts/install-skills.sh (installs into ~/.claude/skills, ~/.codex/skills, ~/.config/opencode/skills), or install the all-in-one Claude Code plugin.

Configuration¶

No key, no quota. The server uses youtubei.js to talk to YouTube's internal InnerTube API, the same endpoint the site itself uses.
Transcript fallback chain: ANDROID-client timedtext track, then the InnerTube transcript endpoint (retried with backoff when it 400s), then local yt-dlp if installed. Each response reports which source served it.
Run it locally. YouTube aggressively rate-limits datacenter IPs - this is a local stdio server by design, not a hosted service.
npx always pulls the latest version, so YouTube internals changes are handled by updating.

Business Relevance¶

Market researchers and product operators mine niches for validated product ideas with demand signals plus pains real builders describe on camera.
Content and SEO strategists find demand-vs-supply gaps: heavily searched topics served by weak, old or misfit videos.
Competitive analysts read a channel's strategy from its own numbers: cadence, outliers, what performs versus what gets published.

Integration with CorpusIQ¶

Complementary to CorpusIQ's business-data connectors: an agent can pull a company's own revenue, bookings and customer context from CorpusIQ, then use TubeScout to mine YouTube for the market signals around it - competitor playbooks, demand gaps and niche validation. CorpusIQ answers what is happening inside the business; TubeScout answers what the market is saying on camera.

Limitations¶

Local-only by design: YouTube rate-limits datacenter IPs, so there is no hosted deployment.
Videos with captions disabled cannot be transcribed (the error says so explicitly).
Caption scraping sits in YouTube ToS gray area - fine for local research tooling, not for building a hosted paid product on it.
Brand new project (npm v0.1.1, repo created Aug 27, 2026, 0 stars) - the fallback chain and rate-limit behavior have not aged in production.