Skip to main content
Glama
GSC Maestro MCP Server

GSC Maestro MCP Server

Application preview

Actual published-source app captured in an isolated browser with built-in illustrative sample data. No connected websites or client properties.

Bring your own accounts and credentials. Nothing in this GitHub repository contains a developer's Google, Bing, Yandex, or ChatGPT access token, connected websites, search data, or private deployment configuration.

Related MCP server: SearchLink Lite

GSC Maestro MCP on Glama

Glama release v0.1.0 was officially published October 27, 2026. Glama's container build and MCP inspection passed, exposing four read-only tools: capability disclosure, CTR calculation, reporting-period comparison, and search-metric explanations. Its tool definition quality score was 4.0/5 across all four tools at publication.

This restricted metrics-only hosted edition operates exclusively on non-sensitive, user-supplied numeric aggregates. It cannot query Google Search Console, Bing, Yandex or client properties, and has no OAuth credentials, local SQLite database, or desktop access. Use your separately installed and authorized GSC Maestro desktop/MCP server for actual provider data.

Features

Area

Capabilities

Google Search Console

Desktop OAuth, authorized properties, queries, pages, daily performance, search opportunities, indexing/URL inspection, sitemap status.

Bing Webmaster

Connect an existing Bing Webmaster API key, sync verified sites, and inspect engine-specific performance.

Yandex Webmaster

Connect a personal Yandex OAuth token, discover properties and query/indexing information when the API permits access.

Technical audit

Bounded public crawler and site-audit summaries with evidence and source attribution.

AI Search and reports

Import supported Bing AI citation exports and generate source-labelled reports.

MCP

Local stdio server, MCP Apps visual dashboard, and optional separately configured bearer/OAuth-protected HTTP gateway.

Offline/local

SQLite snapshots and exports in each user's private application directory. No account credentials bundled with the app.

Platform: Linux desktop (Tauri 2 + React + TypeScript + Rust), plus a Node.js 22 MCP server. Source code is supplied; prebuilt packages and live cloud integrations are not part of this repository. Features dependent on external APIs require those services and their permissions. Yandex API availability in particular is provider-dependent; a configured token is not a guarantee that all reports work.

Build and run locally

Requirements: Node.js 22+, npm, Rust/Cargo, and the system dependencies for Tauri 2 on Linux (notably WebKit2GTK 4.1, GTK3 development headers, and standard Linux build tools). See Tauri's Linux prerequisites.

git clone https://github.com/haseebgb92/GSC-Maestro.git
cd GSC-Maestro
npm ci
npm run desktop

Build an installable Linux Debian package:

npm run desktop:build

Run the local, read-only MCP server from the project folder:

npm run mcp:bundle
npm run mcp

Run the frontend browser preview (without native Tauri account features) using npm run dev. The desktop app is needed to store Google/Bing/Yandex credentials and execute native sync commands.

First-time setup — your own credentials

Google Search Console

  1. Open your own Google Cloud Console, select or create a project, and enable the Google Search Console API.

  2. Set up Google's OAuth consent screen; use the appropriate publishing status/test users for your deployment.

  3. Create an OAuth client ID → Desktop app, then download the client JSON. Do not commit this file or share it with others.

  4. In GSC Maestro → Settings, choose Import OAuth JSON. The app saves the OAuth client settings in the operating-system keyring, not in this repository.

  5. Click Connect Google, authorize the app in your system browser using its read-only Search Console scope, and then refresh/sync your own verified property.

Your Google authorization and refresh token are private to your Linux account. The app does not ship with preconnected properties or the repository author's OAuth client.

Bing Webmaster

  1. Sign into your own Bing Webmaster Tools.

  2. Obtain your API key from the account's API access settings.

  3. Open the app's Bing section, paste the key into the password field, and choose Connect and sync.

The key is saved privately in the user's local application directory (mode 0600). It is never checked into Git, and no built-in key is supplied.

Yandex Webmaster

  1. Obtain a Yandex OAuth token authorized for the Webmaster API through your own Yandex account and OAuth application.

  2. Open the app's Yandex section and configure the token.

  3. Refresh or sync verified sites. API permission, endpoint availability, and Yandex-side restrictions determine which data can be retrieved.

Yandex tokens are stored locally with private permissions. No Yandex OAuth application JSON, tokens, or account IDs from development are included.

Is Google API access paid?

GSC Maestro does not require a paid LLM/API subscription or the Advertpreneur/E92 account. Google Cloud, Bing and Yandex independently govern their APIs, quotas and authorization conditions; check their current terms for your own usage.

Data protection

  • Google: client configuration and refresh tokens are stored in the Linux Secret Service/keyring by the native app.

  • Bing/Yandex: their respective API credentials are stored in per-user private local files under ~/.local/share/com.advertpreneur.gscmaestro/.

  • Analytics: local SQLite databases, report exports, OAuth sessions, logs, cache, credentials, and downloaded files are Git-ignored.

  • MCP: stdio does not expose a public network port. Optional HTTP listens on 127.0.0.1 and creates its own strong local authentication token. Never expose the loopback service directly to the public Internet.

  • No Google, Bing, or Yandex account is connected when a new user first clones this repository.

Optional hosted MCP / ChatGPT OAuth

The public HTTP/MCP endpoint is off by default. Running the local MCP server requires no public URL. If you deliberately want remote access, configure a real HTTPS reverse proxy and an independent hostname belonging to you, then set:

export GSC_MAESTRO_PUBLIC_BASE_URL='https://your-domain.example/gsc-maestro'
npm run mcp:http

Set the variable in your private service environment for unattended deployments; do not hard-code URLs, keys, tokens or OAuth files. .env.example lists optional settings, but it is a template, not an automatically loaded config or a safe place to commit filled-in values.

  • Without GSC_MAESTRO_PUBLIC_BASE_URL, hosted OAuth routes are disabled; private localhost MCP can still use the generated local bearer token.

  • The OAuth implementation's remote client support is presently ChatGPT-specific (PKCE with its known redirect/client metadata). Check your own compatible ChatGPT connector support before promising remote usage.

  • Remote exposure is optional and requires TLS, routing, owner approvals, verification, and normal security maintenance. No third-party tenant access is enabled by publishing this repository.

See SECURITY.md for credential-handling and responsible reporting guidance.

Test and verify

npm run security:check
npm test
npm run build
npm run mcp:bundle
node --no-warnings --experimental-sqlite tests/mcp-smoke.mjs

npm run desktop:build builds the Tauri Linux package. The MCP bundle build regenerates required deployment assets under the ignored src-tauri/resources/ directory.

Source layout

Path

Purpose

src/

React desktop dashboard, charts, Google/Bing/Yandex views, indexing, reports.

src-tauri/src/

Rust native commands, Google OAuth/keyring, private local storage, schedulers.

mcp/

Search adapters, public-site crawler, authenticated HTTP transport, OAuth and MCP server.

mcp-ui/

Visual MCP Apps views.

tests/

Deterministic regression tests; no real credentials required.

src-tauri/icons/

Application branding for packages.

Important limitations

Google Search Console, Bing and Yandex APIs may limit data or require separately verified ownership. Real Google Search Console authorization needs the desktop application and your own OAuth configuration. Search engine metrics are kept separate; sample data is visibly identified and should not be represented as live data. Direct write access to websites is not enabled by these read-only search connectors.

Never open a GitHub issue containing API keys, OAuth JSON, bearer tokens, user property exports, or search-console screenshots with sensitive client data.


Made by Advertpreneur. License: MIT.

Available Tools

4 tools
maestro_calculate_ctrCalculate CTR from supplied search aggregatesA
Read-onlyIdempotent

Calculate clicks divided by impressions, as a percent. All numbers are supplied in the request and must represent the same search source, period and filters. No external queries or stored credentials.

Inputs: clicks (required), impressions (required).

maestro_compare_periodsCompare user-supplied search reporting periodsA
Read-onlyIdempotent

Compare two periods using caller-provided click and impression totals and calculate CTR percentage-point changes. No saved search properties, API calls or estimated traffic.

Inputs: currentClicks (required), previousClicks (required), currentImpressions (required), previousImpressions (required).

maestro_explain_search_metricsExplain search performance metrics and caveatsA
Read-onlyIdempotent

Explain clicks, impressions, CTR, and average position, including why metrics from different engines should not be combined naively. No account access or source data.

Inputs: no named parameters.

maestro_public_capabilitiesPublic GSC Maestro edition: capabilitiesA
Read-onlyIdempotent

Explain what this hosted metrics demo does and cannot do. It cannot access any real Google, Bing, Yandex, or local desktop data.

Inputs: no named parameters.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 4 tools

    Disambiguation4/5

    The two computational tools (maestro_calculate_ctr, maestro_compare_periods) are clearly distinct, and the two explanatory tools differ in subject matter (demo scope vs. metric definitions). There is mild overlap between maestro_public_capabilities and maestro_explain_search_metrics, since both are descriptive/explanatory with no side effects, but the descriptions make the boundary readable.

    Naming Consistency4/5

    All tools share the maestro_ prefix and snake_case naming, which is highly predictable. The only deviation is that three names start with a verb (calculate, compare, explain) while maestro_public_capabilities is a noun phrase, a minor inconsistency.

    Tool Count4/5

    Four tools is on the thin side but fits a deliberately narrow, credential-free metrics demo: two calculators and two explainers. Each tool earns its place and nothing appears redundant, though the surface leaves little room for variation.

    Completeness3/5

    The demo covers CTR calculation and period comparison plus explanatory context, but there is a gap: average position is explained yet no tool computes it, and there is no tool for aggregating or deriving other metrics from supplied clicks/impressions. These are notable but minor omissions given the stated educational scope.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      B
      maintenance
      Read-only MCP server for Google Search Console, with browser-based OAuth flow and local token storage, enabling querying search analytics, site lists, and URL inspection.
      9
      62 PyPI
      MIT
    • A
      license
      A
      quality
      A
      maintenance
      Enables local, read-only access to Google Search Console data, allowing users to query site performance, identify ranking opportunities, inspect URL indexing, and review sitemap status through MCP clients.
      8
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables read-only access to Google Search Console, Bing Webmaster Tools, and PageSpeed Insights, allowing queries for analytics, sites, sitemaps, inspections, crawl issues, and page speed analysis via natural language.
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Connects AI assistants to a site owner's search performance data, letting them ask in plain language about clicks, queries, pages, indexing and sitemaps and get back structured results. Runs locally with your own credentials, read-only by default, and validates requests up front so the API's silent empty-result and terse-error quirks are caught before they mislead.
      22
      95 PyPI
      10
      MIT