SEO Audit Console
github.com/houtini-ai/seo-audit- Category
- Developer Tools
- Rank
- No. 1307Tools index
Previous survey · No. 1299 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Business & Commerce
- Interfaces
- MCP
- GitHub
- 104 stars
- Latest release
- v1.0.6
- Date
About
A free MCP server for Claude that captures and stores the raw data for a technical SEO audit. It merges Google Search Console data, a first-party crawl, DataForSEO and Majestic SEO into one prioritised audit that users can query conversationally, with analysis run from a local SQLite database.
At a glance
SEO Audit Console joins a site crawl with Google Search Console performance data inside an MCP workflow. It is worth trying when a list of crawl errors gives you little guidance about which fixes deserve engineering time.
Good fit
Site owners who already have Search Console data and want to prioritize technical issues with traffic evidence.
Consider something else when
Use a simpler crawler when you only need a one-off link check or lack access to the site’s Search Console property.
Start here
Audit a small site you control. Pick one flagged URL and compare the crawl evidence and traffic figures with what you can verify directly before applying a suggested fix.
Cost and requirements
The Apache-licensed server requires Node 22 or newer, an MCP client and Search Console credentials for the core workflow. Competitor and backlink research can use optional paid data providers.
Limits to know
Impact estimates depend on the crawl and available Search Console history. Local storage is separate from any data your MCP client sends to its model provider.
Sources checked 2026-11-05. Source-based assessment.
Read the source review
What it does
SEO Audit Console is an MCP (Model Context Protocol) server that runs on your own machine and turns Claude into a technical SEO auditor. It pulls your Google Search Console history, crawls your site with its own polite crawler, and keeps both in a local SQLite file per property, joined on one normalised form of each URL. Every finding is then priced by the clicks a fix could recover against the developer hours it costs, so the list arrives in payoff order. Paid market data from DataForSEO and Majestic is optional and runs only when you ask.
Why it's ranked here
The engineering backs up the claims in its own docs. Every finding is tagged as deterministic, drawn from Google's own data, or a judgement heuristic, and the heuristics stay off unless you request them. The crawler follows the current robots standard, caps page reads at 5 MB, backs off when a host rate limits it, and records redirects that leave your site instead of following them. Paid calls are quoted before they run and cached for 20 days. Apache 2.0, published on npm, with a test command that chains about thirty probe scripts over saved fixtures.
What's good
Ranking findings by recoverable clicks per developer hour fixes the classic audit PDF sorted by how alarming each check sounds. The docs show the care: page-experience findings, such as an oversized image, stake a tenth of the page's clicks rather than all of them, a stated assumption that stops one heavy file outranking a real traffic problem. Unknown values stay unknown instead of reading as zero. A Search Console sync writes all or nothing, so a failed pull never leaves half-filled days. Fetches of third-party pages refuse private network addresses, so a hostile redirect cannot reach internal services.
Tradeoffs
Setup is the main cost. You need Node 22 or newer and a Google Cloud service account whose email is added as a user on each property; the docs call that the step everyone misses. The one-click desktop bundle ships native modules built for Windows only, so macOS and Linux users wire it up through npx. The dependency list includes a local machine-learning runtime and an optional headless browser, which is heavy for an MCP server. Competitor, keyword volume and backlink work needs paid DataForSEO, with backlinks on a separate subscription. It crawls sites you own, nothing else.
How to use it well
Best for an in-house SEO or a developer who owns the site and its Search Console property, ideally in Claude Code, where the same session holds the site's repository and can apply the generated redirects, structured data and internal links. Start with the free core: a refresh, then an audit, and leave the judgement checks off until the deterministic list is clean. Keep the browser dashboard on the local loopback address; bound to all interfaces it answers any request that claims to be localhost. Agency work on sites you cannot verify in Search Console is out of scope.
Technical notes+
package.json declares three binaries (seo-audit-console, seo-audit-dashboard, seo-audit-runner), engines node >=22.12.0, and runtime dependencies on better-sqlite3, cheerio, @huggingface/transformers, the MCP SDK and ext-apps, with playwright optional; its test script builds the server and runs a smoke script plus 30 probe scripts, which ARCHITECTURE.md says replay saved fixtures and never call a paid API. ARCHITECTURE.md documents one WAL-mode SQLite file per property joined on url_key, a hand-rolled fetch plus cheerio crawler (Crawlee was rejected because it logs to stdout and would corrupt the stdio channel), RFC 9309 robots handling, readCapped body limits (HTML 5 MB, sitemaps 50 MB inflated, robots 500 KiB), AIMD (additive-increase, multiplicative-decrease) rate-limit backoff, publicOnlyAgent for third-party URLs, a job registry with an external runner for the Docker MCP gateway, and a dashboard webServer with a Host-header DNS-rebinding guard and an HMAC identity check. manual/checks.md gives the priority formula (T x Y x C) / E and the 10% page-experience stake share. src/server.ts reads SERVER_VERSION from package.json and versions its ui:// widget URIs so hosts cannot pin a stale cached bundle.
Observed
- License
- Apache-2.0
- Language
- TypeScript on Node.js, version 22.12 or newer
- Interface
- MCP server over stdio, plus a local browser dashboard and a standalone job runner
- Install surface
- npm package run through npx; a one-click Claude Desktop extension bundle for Windows
- Storage
- One local SQLite database per Search Console property
- Required credential
- Google Cloud service account added as a user on the Search Console property
- Optional paid services
- DataForSEO (keywords, SERP, Labs, backlinks) and Majestic Trust Flow
- Optional free service
- Chrome UX Report API key for field Core Web Vitals
- Test approach
- Probe scripts run against saved fixtures by the npm test command, with no paid API calls
- Dashboard binding
- Loopback address by default, with a localhost Host-header check
Read from README.md, package.json, ARCHITECTURE.md, manual/getting-started.md, manual/checks.md, manual/tools.md, manual/dataforseo.md, .env.example, src/server.ts.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.