Skip to content

smolevich/laviolab-atlas

v0.3.0MIT

Get a side project found: run existing SEO and AI-search audits and order them into one plan, plan content by query clusters, and read a weekly Search Console slice from exports. By LaVioLab.

GitHubOpen Website

LaVioLab Atlas

LaVioLab Atlas

LaVioLab Atlas helps you get a side project found. It routes a live site through search, AI search, content and analytics, and every result links to the page of the Pet Project Atlas that explains why it matters. It is made by Stanislav Shupilkin at LaVioLab and is open source under the MIT licence.

Install

/plugin install laviolab-atlas --marketplace Smolevich/pet-project-atlas

In Codex:

codex plugin marketplace add Smolevich/pet-project-atlas
codex plugin add laviolab-atlas@pet-project-atlas

Commands

CommandWhat it does
/laviolab-atlas:startAsks for your site URL and audience language, runs the third-party seo-audit and geo-audit skills, merges their findings into one list ordered by what blocks what, and names three things to do this week.
/laviolab-atlas:content-planTurns a project description and an optional Search Console export into query clusters, one page per cluster, queued by money.
/laviolab-atlas:reportThe weekly slice: six numbers from Search Console and your own product data, the delta against the previous period, three conclusions and one action.

What it runs and where it connects

The plugin contains instructions (skills) and one remote, read-only MCP server. It ships no scripts, no binaries and no hooks. The skills run on your machine, in your Claude Code session, and only when you run a command.

MCP server laviolab-atlas at https://atlas.smolevich.com/mcp (Streamable HTTP, no login). Its four tools only read the published atlas: search_playbook returns matching sections with a short quote and a link, get_page returns one page in full, list_sections lists the route, and checklist returns the checklist items for a topic such as robots.txt or llms.txt. It receives the words of your question and nothing else from your session, and it never reads your site.

None of the skills pre-approves a shell command or an MCP tool. Their allowed-tools are Skill and Read for start, Read, Write and WebFetch for content-plan, and Read and Write for report.

  • start looks for seo-audit and geo-audit among your skills, reading their SKILL.md in ~/.agents/skills/, ~/.claude/skills/ and ./.claude/skills/, and runs them on the URL you give it. It never installs anything. If an audit is missing, it prints the npx skills add -g … line and stops; you decide whether to install third-party code. Those audits are maintained by other people (coreyhaines31/marketingskills, zubair-trabzada/geo-seo-claude) and are not part of this plugin.
  • content-plan reads files you hand it, such as a query export, and may fetch public pages you name. It writes the plan as a markdown file in your project, so you can commit it next to the pages.
  • report works only from files you download and hand it: the Search Console Performance export with a comparison, the Page indexing totals for submitted pages, and your product's own export. It does not connect to Google, does not drive a browser and never asks for a token or a password. It appends the weekly block to a history file you name.

Data and privacy

LaVioLab Atlas collects nothing. It has no telemetry and no analytics. The MCP server gets only the arguments of a tool call, such as the words of a search, answers from the published atlas and stores nothing; the web server in front of it keeps the usual access log (address, time, path, never the request body). The rest of the network traffic is what is listed above: the sites you ask it to check. Full note: https://atlas.smolevich.com/privacy/. Terms of use: https://atlas.smolevich.com/terms/.

Support

Issues and questions: https://github.com/Smolevich/pet-project-atlas/issues.