# Google Search Console MCP 2026: Setup for Claude, Tools & Quotas | Enrich Labs > Google Search Console MCP explained: no official Google server yet, open-source vs hosted, OAuth/service account setup, URL Inspection quotas, and operator prompts. _Source: https://www.enrichlabs.ai/blog/google-search-console-mcp-2026_ --- ## TLDR **Google Search Console MCP** connects AI clients (Claude Desktop, Cursor, Gemini CLI, ChatGPT via hosted connectors) to your Search Console data so you can ask natural-language questions about queries, pages, CTR, position, sitemaps, and (on fuller servers) URL inspection. Unlike [Google Ads MCP](/blog/google-ads-mcp-2026), Google does **not** ship an official Search Console MCP server (confirmed across vendor docs as of early October 2026). Every GSC MCP path today is **third-party code** calling the public [Search Console API](https://developers.google.com/webmaster-tools). Google does publish MCP for some other products (for example Analytics); GSC is still DIY or hosted. This 2026 guide covers: - What GSC MCP is vs the GSC UI and raw API - Open-source vs hosted options - Auth patterns (OAuth desktop client vs service account) - Setup shape for Claude / Cursor - Operator prompts that map to real SEO work - Quotas and failure modes (inspection limits, 25K row caps) - How this fits next to Google Ads MCP, GEO, and Helena **Primary sources:** [Search Console API](https://developers.google.com/webmaster-tools) · open-source examples such as [AminForou/mcp-gsc](https://github.com/AminForou/mcp-gsc) and [ncosentino/google-search-console-mcp](https://github.com/ncosentino/google-search-console-mcp) · setup writeups from [Windsor](https://windsor.ai/google-search-console-mcp/) and community guides **Related Enrich Labs:** [Google Ads MCP](/blog/google-ads-mcp-2026) · [Best AI for SEO & GEO](/blog/best-ai-for-seo-geo-2026) · [GEO complete guide](/blog/generative-engine-optimization-geo-complete-guide-2026) · product [MCP setup](/mcp) · [Helena](/ai-digital-marketing-agent) * * * ## Table of Contents 1. [What Google Search Console MCP is](#what) 2. [Why there is no official GSC MCP (yet)](#official) 3. [GSC MCP vs UI vs API](#vs) 4. [What the Search Console API can return](#api) 5. [Server and connector options](#options) 6. [Prerequisites and auth](#prereq) 7. [Local setup pattern (Claude / hosts)](#setup) 8. [Operator prompts that earn their keep](#prompts) 9. [Quotas, safety, and common failures](#limits) 10. [GSC MCP vs Google Ads MCP](#ads) 11. [When to use an AI marketing agent instead](#agent) 12. [FAQ](#faq) * * * ## What Google Search Console MCP is {#what} [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) lets an LLM host call tools. A **Google Search Console MCP server** wraps the Search Console API so your assistant can: 1. List verified properties 2. Pull search analytics (clicks, impressions, CTR, position) by query, page, country, device, date, and related dimensions 3. Read sitemap status (and on some servers submit or delete) 4. Optionally run **URL Inspection** for one known URL at a time You stay in Claude or Cursor and ask things like "which pages lost impressions week over week" instead of exporting CSVs from the GSC UI. That is the product promise behind **google search console mcp** and **gsc mcp**: organic search ground truth inside the agent loop. * * * ## Why there is no official GSC MCP (yet) {#official} Vendor comparisons (including Windsor's October 2026 check) state clearly: **Google has not published a first-party Search Console MCP server**, even though it has MCP paths for other Google products such as Analytics. Practical implication: Path Who maintains it Risk Official Google Ads MCP Google Versioned with Ads API docs Community GSC MCP Open-source authors You pick a repo, pin a version, watch scopes Hosted GSC connector SaaS vendors OAuth UX, multi-source chat, pricing, data residency Treat every GSC MCP install as **your** security review: Cloud project, OAuth client, service account grants, and which properties the principal can see. * * * ## GSC MCP vs UI vs API {#vs} Path Best for Strength Weakness **GSC UI** Manual SEO, one property deep dive Full reports, screenshots for stakeholders Slow multi-property ops, weak agent loop **Search Console API** Custom dashboards, ETL, bulk jobs Complete programmable surface Eng cost, quotas, query design **GSC MCP (local OSS)** NL analysis in Claude/Cursor Fast SEO workflows, inspect + analytics You own auth and updates **Hosted GSC MCP / connector** Non-eng marketers, multi-tool chat ChatGPT/Claude web, less install pain Trust, feature gaps (some skip inspection) If you need warehouse-grade history beyond API limits, still ETL into BigQuery or your BI stack. MCP is for interactive analysis, not a full data lake. * * * ## What the Search Console API can return {#api} MCP tools are only as good as the underlying API. API surface What assistants can ask Hard limits (Google-documented class) **Search Analytics** Clicks, impressions, CTR, position by query, page, country, device, date, search type, appearance Up to **25,000 rows** per request; Google returns top rows, not every row; roughly **16 months** history **URL Inspection** Index status, last crawl, canonical, mobile/rich result signals for **one URL** About **2,000 inspections/day** and **600/minute** per property (confirm live quotas) **Sitemaps** Submitted sitemaps, errors/warnings, last download; some servers submit/delete Property permissions required **Sites** List properties the credential can access Verification still required in GSC Inspection is **not** bulk Page Indexing report export and **not** a request-indexing firehose. Scope it to known URLs you already care about. * * * ## Server and connector options {#options} ### Open-source local servers Examples that show up repeatedly in 2026 setup guides: Project Notes [AminForou/mcp-gsc](https://github.com/AminForou/mcp-gsc) (`mcp-search-console` / uvx flows) SEO-oriented tool set; Windsor documents roughly **21 tools** including analytics, period compare, inspection, sitemaps (verify current README) [ncosentino/google-search-console-mcp](https://github.com/ncosentino/google-search-console-mcp) Zero-dependency native binaries (Go/C#), service-account auth, search analytics + sites + sitemaps + inspection tooling Other community servers Vary on OAuth vs service account and tool depth Always pin a commit or release tag. READMEs change faster than blog posts. ### Hosted connectors Hosted options (Windsor, Porter Metrics-style remote MCP URLs, and similar) trade install friction for: - OAuth in the browser - Multi-source chats (GSC + Ads + GA4) - Sometimes **no** URL Inspection Pick hosted when non-engineers need Claude web or ChatGPT. Pick local when you need inspection and tight control of keys. ### Packaged multi-tool MCP products Competitor pages such as Ryze's [Google Search Console MCP](https://www.get-ryze.ai/google-search-console-mcp) sell a packaged connector story. Evaluate them the same way: scopes, write vs read, multi-tenant isolation, and whether inspection is real API or marketing copy. * * * ## Prerequisites and auth {#prereq} Before any Claude config paste: 1. **Search Console property** you can open in the UI (domain or URL-prefix). 2. **Google Cloud project** with the **Search Console API** (Webmasters API) enabled. 3. Auth one of: - **OAuth 2.0 Desktop client** (common for personal Claude Desktop) with client secrets JSON path - **Service account** JSON key, with the service account email added as a user on the GSC property 4. An MCP-capable host (Claude Desktop, Claude Code, Cursor, Gemini CLI, etc.). 5. Full filesystem paths in host configs (Claude Desktop often does not inherit your shell PATH). Never commit client secrets or service-account keys to public git. * * * ## Local setup pattern (Claude / hosts) {#setup} Exact JSON differs by server. Conceptual pattern for OAuth + uvx style packages: - Host key under `mcpServers` (name it `gsc` or `gscServer`) - Command: full path to `uvx` (or the native binary) - Args: package name such as `mcp-search-console` - Env: path to OAuth client secrets file, or service-account file env vars Claude Code one-liner patterns in community docs look like: add MCP with env pointing at `GSC_OAUTH_CLIENT_SECRETS_FILE`, then run the package via `uvx`. Native binary pattern (ncosentino-style): - Download the prebuilt binary for your OS - Pass `--service-account-file` or `GOOGLE_SERVICE_ACCOUNT_FILE` - Add the service account email in Search Console, Settings, Users ### Smoke test checklist 1. Host lists GSC tools after restart 2. "List my Search Console properties" 3. "Top 20 queries by clicks last 28 days for \[property\]" 4. "Pages with impressions over 500 and CTR under 2%" 5. Optional: inspect one known URL and explain index status 6. Confirm the model does not invent properties you do not own Prefer the live README for the repo you install over any static blog snippet. * * * ## Operator prompts that earn their keep {#prompts} ### Coverage and health - "List verified properties and which ones have sitemap errors." - "Compare last 28 days vs prior 28 for clicks and impressions on the domain property." - "Which countries drive the most non-brand impressions?" ### Query and page opportunities - "Queries in positions 8-20 with 200+ impressions and CTR below site median." - "Pages that lost 30%+ impressions week over week with stable position." - "Query+page pairs where we rank for the head term on two different URLs." ### Technical / index - "Inspect these five URLs and summarize index status and canonical." - "Sitemaps submitted in the last 90 days with warnings." ### Reporting narrative - "Draft a Monday SEO standup: total clicks, top movers, three anomalies, three fix ideas." - "Build a brief for content expansion on the top five mid-position pages." **Rule:** treat multi-URL ranking on one intent as a human decision. MCP can surface the pairs; URL strategy stays editorial. * * * ## Quotas, safety, and common failures {#limits} ### Quotas and product reality 1. Search Analytics returns **top rows**, not infinite long-tail 2. URL Inspection is **per URL** with daily/minute caps 3. Fresh "yesterday" data can lag, same as the UI 4. Domain vs URL-prefix properties answer different questions 5. Hosted connectors may omit inspection entirely ### Safety 1. Chat history contains competitive SEO data; treat it as confidential 2. Service accounts with broad property access are high risk in shared laptops 3. Do not paste refresh tokens into public GitHub issues 4. Separate prod properties from sandbox experiments ### Common failures Symptom Likely cause Fix Tools missing Bad path to uvx/binary Use absolute paths Empty property list Wrong Google account / SA not added Verify Users in GSC 403 on analytics API not enabled or scope missing Enable Search Console API Inspection fails Quota or unknown URL Slow down; confirm URL is known Hallucinated rankings Model invented rows Ask for raw tool output * * * ## GSC MCP vs Google Ads MCP {#ads} Search Console MCP Google Ads MCP Official Google server **No** (third-party / hosted) **Yes** ([Ads MCP guide](https://developers.google.com/google-ads/api/docs/developer-toolkit/mcp-server)) Data domain Organic queries, pages, index Paid campaigns, spend, conversions Typical auth OAuth or service account on Webmasters API OAuth/ADC on Ads API + Cloud project access Enrich deep dive This article [Google Ads MCP 2026](/blog/google-ads-mcp-2026) Run both when you need paid + organic in one planning session. Keep credentials and customer IDs separate in prompts so the model does not mix MCC IDs with `sc-domain:` properties. * * * ## When to use an AI marketing agent instead {#agent} Use **GSC MCP in Claude** when: - SEO or eng already lives in Cursor/Claude Code - You want ad-hoc query/page diagnosis - You are OK owning Cloud projects and keys Use an **AI marketing agent** (Helena / Sam) when: - You need scheduled SEO/GEO hunts, drafts, and multi-channel memory - Non-engineers need an approval queue, not a config JSON - Organic work should compound with ads, social, and content ops Start: [Helena](https://agent.enrichlabs.ai/marketing/register) · [/ai-digital-marketing-agent](/ai-digital-marketing-agent) · product MCP onboarding [/mcp](/mcp) · tool shortlist [Best AI for SEO & GEO](/blog/best-ai-for-seo-geo-2026) * * * ## Implementation blueprint (30 days) Week Goal 1 Cloud project + API enable + one property smoke test 2 Prompt pack: movers, CTR underperformers, mid-position pages 3 Optional URL Inspection batch for top money URLs (stay under quota) 4 Decide: stay local, move hosted, or productize via agent * * * ## FAQ {#faq} ### What is Google Search Console MCP? An MCP server or hosted connector that lets AI clients call Search Console API tools so you can analyze organic search performance in natural language. ### Is there an official Google Search Console MCP server? As of early October 2026, Google does **not** publish a first-party GSC MCP server. You use open-source or hosted third-party servers against the public Search Console API. Always re-check Google's developer docs for changes. ### Is GSC MCP free? Open-source servers are free software; you still need Google Cloud setup and you pay for any hosted connector. Search Console itself remains free. ### Can I use it with Claude? Yes. Claude Desktop, Claude Code, and other MCP hosts can load a GSC server once auth and config are correct. Hosted connectors may also work in Claude web. ### Does it support URL Inspection? Depends on the server. Fuller open-source projects expose inspection; some hosted connectors only cover search analytics. Confirm before you buy or install. ### How is this different from Google Ads MCP? Ads MCP is for paid account data and has an official Google server. GSC MCP is for organic Search Console data and is third-party today. See [Google Ads MCP 2026](/blog/google-ads-mcp-2026). ### Will MCP replace the Search Console UI? No. It speeds analysis and agent workflows. Property verification, some report UIs, and stakeholder screenshots still live in GSC. ### What about multi-URL ranking on one intent? MCP can pull query+page combinations so a human (or agent workflow) can spot multi-URL intent overlap. Strategy decisions stay with your SEO process and editorial judgment. ### Where does Helena fit? Helena is the always-on marketing agent layer. GSC MCP is how power users pull organic data into LLM hosts. Many teams use both: MCP for deep dives, Helena for recurring hunts and content ops. * * * ## The bottom line **Google Search Console MCP** is the practical way to put organic ground truth into Claude, Cursor, or a hosted assistant in 2026, even without a Google-official GSC server. Choose open-source when you need inspection and control; choose hosted when marketers need zero-install multi-source chat. Respect Analytics row caps and Inspection quotas, pin server versions, and keep Ads MCP on a separate credential path. For paid-side MCP use [Google Ads MCP 2026](/blog/google-ads-mcp-2026). For SEO/GEO tool selection use [Best AI for SEO & GEO](/blog/best-ai-for-seo-geo-2026). For productized agent ops, use [Helena](https://agent.enrichlabs.ai/marketing/register) and [/mcp](/mcp). **Related:** [GEO complete guide](/blog/generative-engine-optimization-geo-complete-guide-2026) · [Google Ads audit checklist](/blog/google-ads-audit-checklist-2026) · [Best MCP for Meta Ads](/blog/best-mcp-for-meta-ads)