seo-autopilot
An open-source SEO team for any website. It runs in your own GitHub repository with your own Google and AI keys:
- it audits the site and reads Search Console;
- it tracks rankings, researches keywords from your real queries and finds competitor gaps;
- it checks Core Web Vitals, structured data and how AI assistants cite you.
38 specialist agents then turn findings into pull requests. A person approves them, or, if you turn autopilot on, routine changes merge after every automated check passes. It runs on GitHub Actions, so no computer has to stay on, and everything it learns is kept in your own repository.
It is a command-line tool (Node 20+, no dependencies) plus ready-made GitHub workflows. MIT licensed. No telemetry: nothing is ever sent to the maintainers.
Ways to use it
| How | Command | Good for |
|---|
| Full setup in your website's repository | npx github:bishal-paul-git/seo-autopilot#v2 init --url https://example.com/ | the whole team: daily, weekly and monthly jobs, approvals, autopilot |
| One job as a GitHub Action | uses: bishal-paul-git/seo-autopilot@v2 with job: audit (or rank, keywords, schema, …) | adding a single check to an existing workflow |
| Claude Code plugin | /plugin marketplace add bishal-paul-git/seo-autopilot then /plugin install seo-autopilot@seo-autopilot | the 38 agents and the MCP server inside Claude Code |
| Claude or ChatGPT | seo-autopilot mcp locally, or the hosted connector (cloud/connector) | asking questions about your reports in plain words |
init creates a config file, a few editable data files and the GitHub workflows. Then follow docs/getting-started.md, which takes about 15 minutes, mostly on Google's own screens. Once the package is on npm, npx seo-autopilot init works too.
How it compares with paid suites such as Ahrefs and Semrush: docs/compare.md. Common questions: docs/faq.md.
What it does
| Job | When | What you get |
|---|
| Measure | daily | every day | Search Console totals, URL Inspection of the next 60 sitemap URLs, near-miss queries, no-click pages |
| Index | indexnow | every day | up to 10 URLs sent to Bing, Yandex and others |
| Index | indexing-queue | weekly | an issue with the 10 pages to Request indexing for by hand, each with the full URL to paste into Search Console |
| Research | research | weekly | keyword backlog from real queries (near-miss, question, rising, uncovered), up to 2 topic proposals |
| Trends | news | every day | ranks items from your RSS/Atom feeds, proposes at most one topic a day |
| Analyse | digest | weekly | week-on-week search changes, index status, pipeline, backlinks, GA4 sessions, next actions |
| Fix | propose | every day | issues for weak titles and undiscovered pages |
| Write | seo-implement workflow | when you add the label approved | Claude drafts the change on a branch and opens a pull request |
| Publish | schedule + release | daily | approved pages get release dates, N a day, and the site is rebuilt when each is due |
| Links | backlinks, backlinks-queue, mentions | daily / weekly | capped, careful submissions; a hand-done checklist; unlinked brand mentions with a drafted request |
| See | dashboard | every day | one HTML file with charts, index status, the schedule, keywords and backlinks |
| Manage | ui (local app) | on demand | one app on your computer for every project: dashboard, approvals from GitHub, sync, dry runs |
| Keywords | keywords | weekly, or on request | keyword explorer: topics, intent, demand, difficulty, gaps no page owns, questions |
All of the research and site tools run on your own Search Console data first. Each key you add (most are free) switches on more: see docs/research-suite.md.
The one rule: checks approve routine work, a person approves risky work
- By default nothing publishes without you. The tool opens an issue ("Topic: …") with the evidence. When you add the label
approved, the implementer drafts the change on a branch as a pull request. Nothing is live until it is merged.
- Autopilot is opt-in (repository variable
SEO_AUTOPILOT=true, decision D17). Routine changes are merged automatically, but only after every automated check (build, facts, reviewer) has passed on that exact commit. Routine means fixes, titles and descriptions, internal links, refreshes, and a few new pages a week. The pace limits apply, and the seo-watch job undoes a change that makes things worse.
- Always a person: sensitive topics (health, money, legal), site-wide indexing changes, migrations, deletions, outreach, and changes to workflows or dependencies.
SEO_AUTOPILOT_PAUSED=true stops everything.
- A merged page is marked queued. The scheduler dates it (default 3 a day), and the site goes live with it on that day.
- Hard rules baked into every prompt: no invented numbers, clients, quotes, prices or results; cite real sources.
Read docs/content-automation.md for the full picture.
Honest limits
- Google indexing cannot be automated. The Indexing API is only for job and livestream pages. The tool prepares the list; you press the button (about 10 a day).
- IndexNow reaches Bing, Yandex, Seznam and Naver, not Google.
- Keyword volumes are not included by default. Research uses your own Search Console impressions. Estimated volumes and competitors' rankings need your own DataForSEO key, and are labelled as estimates. AI "ideas" are flagged as unverified.
- The paid-provider integrations (DataForSEO, SERP providers, AI engines) and the free Google APIs have been tested with recorded responses, not yet against the live services. Check a report against the provider's own dashboard the first time you use a key.
- Backlinks cannot be fully automated. Most free directories need an account or show a CAPTCHA, and this tool never solves CAPTCHAs, never creates accounts, never pays and never swaps links. See docs/backlinks.md.
- Nothing here guarantees rankings or indexing.
Install options
- Per project (recommended):
npx github:bishal-paul-git/seo-autopilot#v2 init --url … inside each website repo. The generated workflows call the same version tag, so you get fixes by moving the tag, or pin a version like #v2.0.0.
- Several sites from one repo: add more entries to
sites in the config; jobs run for each site.
- As a Claude Code plugin:
/plugin marketplace add bishal-paul-git/seo-autopilot, then ask Claude to "set up SEO for this site". It follows the same steps.
Commands
seo-autopilot init --url https://example.com/ scaffold config, data files and workflows
seo-autopilot doctor check the setup and say what to fix
seo-autopilot run <job> [--site id] [--dry] run one job
seo-autopilot ui [--open] local app to manage every project on this computer
seo-autopilot mcp [--project folder] MCP server for Claude, ChatGPT and other AI assistants
seo-autopilot projects add|list|remove [folder] choose which website folders the app shows
seo-autopilot state restore|save load or save memory on the "seo-data" branch
Documentation
- Getting started
- Connect Google: Search Console and Analytics
- Content automation and scheduling
- Dashboard
- The local app
- The expert layer: audit, insights, priorities, experiments
- Research suite and site intelligence: keywords, rank, gap, optimiser, AI visibility, authority, link graph, vitals, schema, entity
- Use it from Claude or ChatGPT (MCP)
- Backlinks
- Configuration reference
- Jobs and secrets
- Measurement and compliance: consent, sensitive topics, pace, citations
- The agents
- Python and Go engines
Landing page
The product site lives in site/ (with a sample-data demo of the dashboard) and deploys to GitHub Pages from .github/workflows/pages.yml once Pages is switched on for the repository.
Status
Version 2 is used in production on the maintainers' own sites. Contributions and bug reports are welcome: see CONTRIBUTING.md.
