On this page7
Connect YouTube Analytics to Claude in two ways: add Otto's hosted HTTP MCP server (https://otto.islaintel.com/api/mcp) with a single otto_mcp__ token after a one-time Google OAuth login in Otto, or self-host a local MCP server against the YouTube Analytics and YouTube Data APIs. Both paths keep YouTube strictly read-only via the https://www.googleapis.com/auth/youtube.readonly scope required by Google's YouTube Analytics API. Pasting wide YouTube Studio exports into Claude burns context on every turn; querying aggregated reports.query rows through MCP returns only the metrics you ask for. Separately, Otto documents that uploading a raw 3,000-row Ads Manager CSV can consume 40,000 to 80,000 tokens in a single Claude turn, whereas querying aggregated metrics through Otto MCP cuts token usage by 80β90% across 227 tools and 15 integrations.
Copy-paste Claude prompts to test your YouTube Analytics MCP connection
Paste these five prompts directly into Claude once your YouTube account connects through the Model Context Protocol.
Prompt 1: 30-day channel and video performance audit
Query my YouTube Analytics data for the last 30 days. Pull total views, estimated minutes watched, average view duration, and net subscribers gained. Break down the top 10 videos by watch time and flag any video where the average view duration dropped below 35%.
Prompt 2: Traffic sources vs. Search Console queries
Pull the traffic sources report for my top three videos over the past 60 days. Group views by YouTube search, suggested videos, external traffic, and direct links. List the YouTube search terms that drove the highest average watch time, then pull my Google Search Console queries for the last 28 days. Flag queries where YouTube sends traffic but organic search impressions are flat or falling.
Prompt 3: Cross-channel organic video audit
Compare my YouTube video performance over the last 30 days with my organic reach on connected Instagram and Facebook Pages. Which topics produced the highest engagement on socials, and do we have corresponding long-form videos on YouTube covering those themes?
Prompt 4: Video-to-site funnel with GA4 and PageSpeed
Using connected YouTube Analytics, list my top five external referral domains and the landing pages they hit over the last 30 days. Cross-check those URLs in GA4 for sessions, bounce rate, and conversions. For each landing page, run psi_analyze_url and report Lighthouse performance and Core Web Vitals. Summarize which videos send traffic to slow pages.
Prompt 5: WordPress draft and paid amplification
Identify my five highest-retention YouTube videos from the past 90 days. For the top-performing video, extract the core arguments and draft a WordPress post in draft status. Then find the two videos with the highest retention from non-subscribed viewers and outline paused Meta Ads campaign angles and DRAFT LinkedIn Ads tests for those concepts.
Pair YouTube winners with paid tests using 10 Claude workflows for Meta Ads in 2027 and how to connect Claude to Meta Ads in 2027.
Doing it by hand in YouTube Studio vs. DIY Google Cloud MCP vs. Claude + Otto MCP
Querying channel metrics through an active protocol changes how fast Claude returns insights and how many tokens remain in your context window.
| Dimension | Manual YouTube Studio CSV export | DIY Google Cloud local MCP | Claude + Otto MCP |
|---|---|---|---|
| Setup time | Zero setup; repeated manual downloads | Varies by team; Google Cloud OAuth, API enablement, and a local MCP server | One-time Google OAuth inside Otto; no client-secret JSON in Claude |
| Claude token usage | Wide Studio CSV rows fill the context window each paste | Aggregated JSON per query (you manage quota) | 80β90% lower token usage vs 3,000-row Ads Manager CSV uploads; aggregated YouTube queries instead of wide exports |
| Credential storage | Raw CSV files on local disk | Client secret JSON files stored on local disk | Server-side OAuth; Claude client holds only an otto_mcp__ token |
| API maintenance | None; manual UI exports only | You manage token refreshes, OAuth consent, and quotas | Hosted HTTP endpoint maintained by Otto |
| Multi-channel scope | YouTube data only | YouTube data only | 227 tools across 15 integrations in Otto MCP |
| Downstream action | Manual copy-pasting to website or ad accounts | None; inspection only | Previews GitHub pull requests and WordPress drafts |
Otto MCP
Ask your AI assistant what changed in your ads
Connect Claude or ChatGPT to Otto, then ask about Meta Ads, Google Ads, GA4, and Search Console in the same chat. The logins stay in Otto.
How the YouTube Analytics API (reports.query) and YouTube Data API work under the hood
Querying YouTube through an AI assistant relies on two distinct Google APIs.
The YouTube Analytics API handles private channel performance. Its primary method, reports.query, retrieves metrics such as views, estimatedMinutesWatched, averageViewDuration, and subscribersGained. Google requires API requests targeting channel data to specify the https://www.googleapis.com/auth/youtube.readonly OAuth scope, as detailed in Google's YouTube Analytics API documentation. To retrieve channel metrics, the API request must set the ids parameter to channel==MINE or supply an authorized channel identifier, according to Google's Channel Reports documentation.
The YouTube Data API v3 handles channel structure and metadata. It retrieves video titles, descriptions, tags, playlist IDs, and upload dates through endpoints like channels.list and videos.list. Projects that enable the YouTube Data API receive a default quota of 100 search.list calls, 100 videos.insert calls, and 10,000 units per day combined for all other endpoints, as documented in Google's YouTube Data API overview. A read operation that lists channels, videos, or playlists usually costs 1 unit; a write usually costs 50 units.
Claude Request
β
βΌ
MCP Protocol Gateway (HTTP or stdio)
β
βββ> YouTube Data API (channels.list, videos.list) ββ> Titles, tags, upload dates
β
βββ> YouTube Analytics API (reports.query) βββββββββ> Watch time, retention, traffic sources
Anthropic's Model Context Protocol standardizes how models query these endpoints. Instead of processing 50 columns of raw CSV export data, Claude calls targeted tools that request only the required dimensions. Remote servers connect directly across the network without local runtime dependencies, as described in Model Context Protocol's remote server documentation.
How to connect Claude to YouTube Analytics in five steps with Otto MCP
Otto Webmaster is an AI webmaster and marketing teammate built by IslaIntel in San Juan, Puerto Rico. Follow these five steps to link YouTube Analytics to Claude through Otto's hosted MCP endpoint.
1. Authorize your YouTube channel inside Otto
Log into your Otto Webmaster dashboard and navigate to Settings -> Integrations -> YouTube. Complete the one-time Google OAuth verification prompt. This step grants read-only access to your channel metrics without exposing Google client secrets to local config files. Detailed setup instructions appear in the connect YouTube to Otto guide.
2. Copy your Otto MCP bearer token
Navigate to your organization settings to generate an MCP token. Otto assigns a secure bearer token prefixed with otto_mcp__. This token scopes Claude's access strictly to your connected organization and integrations. Detailed instructions live in the Claude MCP setup documentation.
3. Add Otto's hosted HTTP endpoint to Claude
Add the Otto MCP endpoint to your Claude configuration. For Claude Desktop, edit claude_desktop_config.json:
{
"mcpServers": {
"otto": {
"type": "http",
"url": "https://otto.islaintel.com/api/mcp",
"headers": {
"Authorization": "Bearer otto_mcp_YOUR_TOKEN_HERE"
}
}
}
}
Claude Code, Cursor, and ChatGPT connect using this same HTTP URL and bearer token header. Setup steps for each client live at connect Claude to Otto MCP.
4. Run a verification query across your marketing stack
Restart Claude and confirm the tools appear in your active session. Ask Claude:
Confirm my YouTube connection by listing my channel title and total views over the last 7 days.
Claude will execute the request and return your aggregated channel metrics. In the same turn, Claude can compare these video numbers against keyword trends in Otto's automated weekly SEO reports or pull GA4 and Search Console metrics through the same MCP session. See Otto Website Analytics & Weekly Briefings for how briefings are delivered in the Otto app.
5. Pair YouTube retention insights with website and ad updates
Once connected, use high-performing video concepts to guide site content and campaigns. When a video demonstrates high viewer retention, prompt Claude to outline a detailed landing page or article. Through Otto AI Website Manager, Claude can draft the piece directly in WordPress or open a GitHub pull request. If you amplify that topic via paid ads, reference how to connect Claude to Meta Ads in 2027 to build out matching ad sets.
Alternative: when to self-host a local YouTube Analytics + YouTube Data API MCP server
Engineering teams requiring standalone YouTube inspection without additional marketing integrations can build a local stdio MCP server.
First, open the Google Cloud Console and create a new project. Enable both the YouTube Analytics API and the YouTube Data API v3. Google assigns new projects 100 search.list calls, 100 videos.insert calls, and 10,000 units per day combined for all other endpoints, as noted in Google's YouTube Data API documentation.
Second, configure an OAuth consent screen. Add the https://www.googleapis.com/auth/youtube.readonly scope, as required by Google's YouTube Analytics API reference. Add your Google account email as an authorized test user while the app remains in testing mode.
Third, generate an OAuth 2.0 Client ID for a desktop application. Download the resulting client credentials JSON file to your local development machine.
Fourth, write a local MCP server script using the Python or TypeScript MCP SDK. Your server must execute the OAuth loop, handle token refresh routines locally, and expose tools that wrap reports.query with ids=channel==MINE, following Google's Channel Reports guidelines.
Fifth, add your script to claude_desktop_config.json under the stdio transport:
{
"mcpServers": {
"local-youtube": {
"command": "python",
"args": ["/path/to/youtube_mcp_server.py"]
}
}
}
This local approach works for dedicated YouTube data queries. However, it requires maintaining local token stores, runs only on the machine where configured, and does not connect to GA4, Search Console, WordPress, or ad networks.
Read-only YouTube vs. draft-first website edits vs. live social publishing: Otto's permission model
Otto enforces strict permission boundaries across connected services to protect ad budgets and production sites.
Permission Model
βββ Read-Only (Inspection Only)
β βββ YouTube Analytics & Data
β βββ Google Sheets
β βββ PageSpeed Insights (psi_analyze_url)
β
βββ Limited Writes (Pro, Reviewer/Admin)
β βββ Google Analytics 4 (key events, custom dimensions)
β βββ Search Console (sitemaps, indexing requests)
β βββ Google Business Profile (review replies, local posts)
β βββ Shopify (new products start as DRAFT)
β
βββ Draft-First (Safe Staging)
β βββ WordPress (29 tools; posts/pages default to draft)
β βββ GitHub (Opens pull requests with live previews)
β βββ Google Ads (Search campaigns start paused)
β βββ LinkedIn Ads (Campaigns created in DRAFT state)
β
βββ Reviewer/Admin Gated (Immediate Execution)
βββ Facebook Pages (Publishes immediately)
βββ Instagram (Publishes immediately)
In Otto MCP, YouTube, Google Sheets, and PageSpeed Insights are strictly read-only. YouTube executes no spend, no video deletions, and no channel modifications. Google Analytics 4, Google Search Console, Google Ads, Google Business Profile reviews, and Shopify allow only limited changes on Pro, each made by a reviewer or admin; turning on a Google Ads campaign or changing a budget needs an explicit confirm. You can run Lighthouse audits on any URL using Otto's platform key via psi_analyze_url without supplying your own Google API key.
For website management, Otto operates draft-first. Otto's WordPress integration provides 29 tools across posts, pages, categories, tags, and media; new posts and pages default to draft status. On GitHub repositories, site updates open as pull requests with preview URLs. Checkout, contact, and legal pages always require explicit human approval.
Paid ads follow similar guardrails. LinkedIn Ads campaigns and ad sets are created as DRAFT by default, rejecting $0 budgets and requiring confirm=true alongside reviewer or admin rights to activate or pause. Meta Ads write operations also require reviewer or admin permissions.
In contrast, publishing updates to Facebook Pages and Instagram through Otto MCP goes live immediately and requires reviewer or admin permissions. You can review setup requirements in the connect Instagram to Otto guide.
Plans, honest limitations, and frequently asked questions
Service businesses, clinics, law firms, and agencies use Otto to manage sites, search, and reporting without hiring a full-time marketer or paying agency retainers for routine execution. Unlike ad-focused tools such as Ryze AI, which concentrates on paid media and SEO without live GitHub pull requests or WordPress drafts, Otto handles landing page updates directly when ad performance shifts. See our breakdown in 7 best Ryze AI alternatives for paid ads and SEO in 2027. Tools like Synter connect over 40 ad networks but do not manage site code or technical SEO. Blaze AI focuses on drafting social and newsletter copy rather than deep diagnostics. Stormy AI concentrates on creator discovery rather than full-funnel operations.
Otto plans cover both core webmaster features and Otto MCP:
- Starter ($9.99/week or $29.99/month, paid in full, no free trial): MCP only: reads ads, socials, and analytics and publishes social posts in Claude, ChatGPT, or Cursor, with up to 1,000 data requests a day and one seat. It can't schedule, change ads, generate images, or edit the site.
- Pro ($199/month, or $99/month billed annually at $1,188/year): Starts with $1 today for 3 days, then the plan price; card required. Full website management, the Otto app and chat, write access to paid ads, post scheduling, and comment replies, image generation, team seats, and WhatsApp, email, and voice. Cancel anytime.
- Enterprise (no published price): A call with the founder, for teams that want Otto across more than one brand.
Full plan details are available on our pricing page.
Honest limitations
- Ad changes require Pro: Starter reads YouTube, ads, and socials and publishes social posts; active campaign creation, scheduling, and comment replies require the Pro plan.
- Agency portal limits: White-label agency portals remain limited.
- Terminal CLI status: The Otto CLI for terminal agents is coming soon; remote MCP is the supported integration path today.
- Catalog boundaries: Unannounced integrations such as Wix, Webflow, and TikTok are not part of the live MCP catalog.
- No ranking promises: Otto structures site schema, listings, and pages so search engines and AI assistants can read the business, but does not promise Google Map Pack placement, search rankings, or ChatGPT mentions.
Frequently asked questions
Can Claude edit or delete YouTube videos through Otto MCP?
No. In Otto MCP, YouTube is strictly read-only via the youtube.readonly scope. Google Sheets and PageSpeed Insights are also read-only. Google Analytics 4, Search Console, Google Ads, Google Business Profile reviews, and Shopify allow only limited changes on Pro, each made by a reviewer or admin.
Does Claude receive my Google account password?
No. You authenticate YouTube once via Google OAuth inside Otto. Claude connects using an organization-scoped bearer token (otto_mcp__), and OAuth secrets never leave Otto's servers.
Can I approve website changes from WhatsApp?
Yes. Otto includes an enhanced WhatsApp integration that allows managers to review, approve, or reject website edits directly from chat without logging into the dashboard.
Otto AI Paid Ads Manager
The ad should match the page.
Otto watches the accounts next to the live site. Creative and budget stay drafts until you approve them. The weekly note says what the spend did.
See how Otto does it