Skip to main content
Glama

EdGEO

EdGEO MCP server

A free, local-first MCP toolkit for evidence-based SEO, GEO, and Google AI Search audits.

EdGEO gives compatible AI agents tools to inspect public website pages, crawl a bounded part of a site, analyze AI answers, and read a user's own Google Search Console data. It is a standalone open-source project. No EdGEO account, hosted EdGEO service, telemetry, or subscription is required.

An API answer is not the same as what a person sees in ChatGPT, Gemini, Google AI Overviews, or another consumer product. EdGEO reports the source and conditions for each observation and does not promise rankings.

What EdGEO can do

  • Audit a page: inspect HTTP status and redirects, title and description, canonical and robots directives, headings, language, viewport, links, image alt text, Open Graph/Twitter metadata, hreflang, and parseable JSON-LD.

  • Crawl a bounded site sample: make up to 50 same-host URL fetch attempts, starting from the requested URL and same-host sitemap URLs. EdGEO preserves trailing slashes when requesting URLs, follows robots.txt and crawl delay, separates non-HTML resources, and deduplicates pages that resolve to the same final URL. Remaining queued URLs are not a sitewide coverage estimate.

  • Collect an AI answer: query an Ollama server on loopback or query the Gemini API after explicit consent. Returned observations include provider, model, prompt type, locale, timestamp, and limitations.

  • Analyze and compare AI answers: analyze supplied or collected answers for literal brand mentions and domain citations. Compare answers only when prompt and recorded conditions match.

  • Connect to Google Search Console: authorize the user's account with a read-only OAuth scope, list available properties, and query selected dates, dimensions, and search types.

  • Import exported data: parse Search Console CSV/TSV locally, including common English and Indonesian headers, or summarize structured rows.

Related MCP server: SEO Screaming Toad MCP Server

Install in an AI agent

Install the local MCP server in selected clients with:

npx --yes --prefer-online edgeo-mcp@latest install --agent claude,codex,cursor,gemini

For a native Gemini CLI extension installation, use:

gemini extensions install https://github.com/edgarcia-id/edgeo

The extension registers the MCP prompt as /edgeo. Its optional Gemini API key setting is only needed for EdGEO's Gemini provider; the key is kept in Gemini CLI's sensitive extension settings.

For the Claude Code marketplace plugin, add the EdGEO marketplace and install its plugin:

claude plugin marketplace add edgarcia-id/edgeo
claude plugin install edgeo@edgeo-marketplace

The plugin provides MCP tools and the namespaced /edgeo:edgeo command. The standalone installer below also installs an unqualified /edgeo skill.

The installer merges user-level settings, backs up JSON config files before editing them, and preserves unrelated server entries. It upgrades earlier EdGEO-generated Cursor and Gemini MCP entries in place. Claude Code and Codex use their native CLI commands; Claude Code and Codex also receive an EdGEO skill. Restart or reload the selected clients after installation.

Automatic updates

The installer configures the MCP server to run npx --yes --prefer-online edgeo-mcp@latest serve. Each time an agent starts the server, npm checks for the latest published EdGEO version and downloads it when needed. A server already running keeps its current code until the agent reloads or restarts it. The update check sends the package name and version request to npm; it does not send the audited website, prompts, API keys, or local results.

Claude Code installs the /edgeo:edgeo plugin from a third-party marketplace, whose automatic updates are off by default. To receive plugin and prompt updates automatically, open /plugin, select Marketplaces → edgeo-marketplace, and enable Auto-update. Claude Code checks marketplace updates at startup and notifies you when a plugin changes; run /reload-plugins or start a new session to load it. A session already using EdGEO keeps the version it loaded.

Tagged GitHub releases publish the matching package version to npm and update EdGEO's Official MCP Registry listing through GitHub Actions. Before the first automated publish, configure edgeo-mcp on npm with a trusted publisher for GitHub Actions, repository edgarcia-id/edgeo, workflow release.yml, and permission to publish. See npm's trusted publishing setup. Keep the tag in the form v<package.json version>.

npx --yes --prefer-online edgeo-mcp@latest doctor
npx --yes --prefer-online edgeo-mcp@latest uninstall --agent claude,codex,cursor,gemini

The MCP server registers a prompt named edgeo. Hosts that expose MCP prompts as slash commands may show /edgeo; other hosts use their skill picker, prompt UI, or normal tool invocation. MCP does not define one slash-command syntax for every agent.

For direct/manual setup and supported host behavior, read Install and distribute EdGEO and agent install notes.

Search Console setup

Google Search Console API usage is free of charge and subject to Google's quotas. Each user connects their own Google account; EdGEO requests only the Search Console read-only scope. To connect, create a Google OAuth client of type Desktop app, enable the Search Console API in that Google Cloud project, and run:

npx --yes --prefer-online edgeo-mcp@latest search-console connect --client-id YOUR_DESKTOP_OAUTH_CLIENT_ID

The browser opens Google's consent screen and returns to a loopback callback on the same computer. The refresh token is stored in the user's local EdGEO data directory; set EDGEO_DATA_DIR to choose another directory. edgeo-mcp search-console status checks local setup and edgeo-mcp search-console disconnect revokes the saved grant and removes it.

See Google Search Console connection details for OAuth setup and data handling.

Portable Agent Skills

EdGEO includes three optional, portable Agent Skills for focused workflows: technical SEO audits, AI visibility research, and Search Console analysis. They follow the open SKILL.md folder format and use EdGEO's MCP tools; install the EdGEO MCP server separately in the agent first. See Agent Skills for their scope and installation notes. The skills are available in the source repository and npm package; each can also be submitted as its own free MCP Rush listing.

Model connections

  • Ollama: local API requests go to 127.0.0.1:11434; no model API bill is added by EdGEO. Local hardware is required. A cloud-backed Ollama model may still transmit prompts externally.

  • Gemini API: configure GEMINI_API_KEY in the MCP process environment. The tool requires consentToExternalRequest: true before sending a prompt. Some model tiers have free quotas; rates, availability, and data terms vary. Do not send confidential data unless the provider's terms and settings are acceptable.

Neither connector automatically performs consumer-app testing or web search. Returned answers have no assumed citations. See model connection details.

Requirements and local development

  • Node.js 20.18.1 or later

  • An agent that supports local stdio MCP for agent use

npm ci
npm test
npm run typecheck
npm start

npm start runs the local stdio MCP server. The server writes diagnostic messages to stderr and reserves stdout for MCP messages. npm test builds the TypeScript and exercises CSV parsing, visibility analysis, API request shaping, OAuth token refresh, installer config merging, SSRF rejection, and MCP initialization/tool discovery.

Method limits

  • Page audits inspect server-returned HTML and headers. They do not render JavaScript, measure Core Web Vitals, verify indexing, or observe search rankings.

  • Crawling is bounded to the exact hostname, obeys applicable robots.txt rules, and stops when robots.txt cannot be read.

  • Search Console only reports fields and appearance labels returned by Google. EdGEO does not infer AI Overviews, AI Mode, or SGE from ordinary web search rows.

  • Brand matching is literal and may need human entity review. A mention is not a positive description or recommendation; a citation does not prove support for a nearby claim.

  • Comparisons are descriptive, not causal. Differences may reflect sampling, model updates, retrieval, account state, or unrecorded conditions.

  • Search Console exports and API responses can omit anonymized queries or be truncated. A click, citation, or impression is not revenue or proof of a GEO effect.

Privacy and security

Page and crawl requests go directly to the public site selected by the user. EdGEO blocks local/private IP targets and pins validated DNS addresses. Supplied data and API results are processed locally and returned to the connected agent; the agent host may keep conversation or tool data under its own policies. Gemini prompts leave the computer only after an explicit tool consent flag. See privacy and data handling and security policy.

License

Apache-2.0. Commercial use and redistribution are permitted subject to the license and included attribution notices. See LICENSE and NOTICE.

Project status

EdGEO is distributed on npm and listed in the Official MCP Registry. Release history: 0.1.4 adds an optional Search Console property filter; 0.1.3 added automatic update checks and expanded the default crawl limit; 0.1.2 refined crawl URL handling, redirects, sitemap observations, and heading cues. Version 0.1.1 added native Gemini CLI extension support; the Gemini gallery discovers qualifying tagged repositories through its daily crawler.

Available Tools

9 tools
edgeo_analyze_visibilityAnalyze supplied AI answersA
Read-only

Analyze caller-supplied answers with transparent brand alias and citation checks. Keeps prompt types and failures separate; does not call models or automatically label recommendations.

Inputs: brandName (required), observations (required), aliases, brandDomain.

edgeo_audit_pageAudit a public webpageA
Read-only

Fetch and audit one public HTTP/HTTPS HTML page for observable technical SEO and page-structure signals. Does not render JavaScript, measure Core Web Vitals, or verify search indexing.

Inputs: url (required).

edgeo_compare_visibility_runsCompare two sets of supplied AI answersB
Read-only

Compare complete answers only when the normalized prompt and recorded source, provider, model, surface, locale, search mode, and prompt type match. Returns paired mention and citation changes with sample counts; does not infer causation or call AI providers.

Inputs: after (required), before (required), brandName (required), aliases, +1 more.

edgeo_crawl_siteCrawl and audit a public siteA
Read-only

Run a bounded same-host HTML crawl seeded from the start URL and same-host sitemap URLs. Defaults to 50 attempted URLs and accepts up to 200. Preserves trailing slashes when requesting URLs, reads robots.txt first, applies its rules and crawl-delay, separates non-HTML sitemap resources, and never renders JavaScript or follows links to other hosts. Reports known queued URL counts, not sitewide coverage.

Inputs: url (required), delayMs, maxPages.

edgeo_import_search_console_csvImport a Search Console CSV locallyA
Read-only

Parse a user-provided CSV or TSV export in memory, recognize common English and Indonesian Search Console headers, and return a transparent summary. The file contents are not uploaded or persisted by this tool.

Inputs: csvText (required), columnMap, delimiter.

edgeo_query_modelCollect one AI model answerB
Read-only

Send one prompt to a configured local Ollama model or, with explicit consent and GEMINI_API_KEY configured, Google Gemini API. Returns an observation ready for EdGEO visibility analysis. Provider API output is not equivalent to a consumer AI app or Google AI Search result; no citations are assumed.

Inputs: model (required), locale (required), prompt (required), provider (required), +3 more.

edgeo_search_console_queryQuery Google Search ConsoleB
Read-only

Query the user's authorized Google Search Console property for web, image, video, news, Discover, or Google News performance rows. Requires edgeo-mcp search-console connect. Reads Search Console directly with OAuth read-only access; AI feature labels are never inferred from generic search rows.

Inputs: endDate (required), siteUrl (required), startDate (required), rowLimit, +3 more.

edgeo_search_console_sitesList connected Search Console propertiesA
Read-only

List Google Search Console properties for the user's locally authorized Google account. Requires prior setup with edgeo-mcp search-console connect; uses a read-only OAuth scope and returns only properties the Google account can access. Optionally filter property URLs by a case-insensitive substring.

Inputs: siteUrlContains.

edgeo_summarize_search_consoleSummarize supplied Search Console rowsA
Read-only

Summarize user-supplied Search Console rows by provided feature label, including clicks, impressions, CTR, and impression-weighted average position. Does not connect to Google or infer an AI feature from unlabeled data.

Inputs: rows (required).

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    A3.7/5.0

    Scored across 9 tools

    Disambiguation4/5

    Most tools occupy clearly distinct slots (crawling, single-page auditing, model querying, Search Console access), and descriptions actively draw boundaries — e.g. analyze_visibility works on supplied answers while query_model produces them, and summarize_search_console vs. compare_visibility_runs differ by input type. The closest pairs are the three Search Console data-entry points (query/search_console_sites/import_csv) and analyze_visibility vs. summarize_search_console, which both digest supplied rows and could be momentarily confused.

    Naming Consistency4/5

    All names are snake_case under a uniform `edgeo_` prefix with a leading verb, which is predictable and readable. Minor deviation: three tools embed the compound noun `search_console` after the verb, so the verb_noun grammar is less clear there (edgeo_search_console_sites reads as a list operation but has no `list_` verb).

    Tool Count5/5

    Nine tools sit comfortably in the well-scoped range, and each covers a distinct step (crawl, audit, query, import, query API, summarize, analyze, compare, list properties). No obvious filler or trivially overlapping entries.

    Completeness3/5

    The surface covers acquisition, analysis, and comparison, but there is no tool to create, persist, or list analysis runs — compare_visibility_runs requires matched completed runs that nothing in this set produces or retrieves, leaving a dead end. There is also no export/report tool to carry findings out, so agents must stitch workflows together externally.

    Maintenance

    ActivityMaintained
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      C
      maintenance
      Lets any AI agent audit any website for SEO, GEO/AEO, and speed problems, reporting issues and recommendations without modifying the site.
      4
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Provides agents with tools to score a page's AI readiness, check which AI crawlers robots.txt blocks, and assess passage citability—all performed locally.
      58 npm
      2
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to crawl live websites, audit AEO readiness, generate Schema.org @graph JSON-LD, llms.txt, ai.txt, and robots.txt, inject structured data into HTML, validate optimizations, and retrieve framework-specific code snippets.
      MIT