Repo of the day: OpenSEO, open source SEO with an MCP server for your AI agent

💡 What it is: a SEO suite you self-host

OpenSEO bills itself, no hedging, as an open source Semrush and Ahrefs alternative: the same SEO-suite workflows — keyword research, rank tracking, competitor analysis, backlinks, technical audit — but as MIT code running on your own infrastructure. No monthly subscription: the app costs zero, and you pay for data on demand through a DataForSEO key you bring yourself.

The stated philosophy is the opposite of the «does everything but it's heavy» suite: focused workflows instead of an overloaded panel. And since the source is open, the project invites you to fork it and vibe-code your own tailored tool — not a random detail, given the whole front-end is TypeScript on Cloudflare Workers, the same stack this blog runs on.

# 1. clone and prepare the variables
cp .env.example .env

# 2. set your data key in .env
#    DATAFORSEO_API_KEY=...

# 3. bring it up with Docker and open the browser
docker compose up -d
#    http://localhost:3001

🧠 The twist: built for you and for your agent

This is why OpenSEO makes our repo of the day. The project isn't just a web app: it exposes an MCP server that puts SEO data at the disposal of any agent — Claude Code, Codex and the like — together with a set of ready-made skills. In practice your agent stops «talking about SEO» and starts querying real keywords, backlinks and SERPs while it works.

It's the difference between asking a model for generic advice and giving it the real data to reason on. A skill like keyword-research or competitor-analysis walks the agent through a repeatable flow: it gathers the numbers via MCP, clusters them, and produces recommendations you can check line by line. SEO moves into the process, it doesn't stay a dashboard to stare at.

🧰 The workflows covered: from keywords to AI visibility

On the feature side OpenSEO covers the items you'd expect from a suite, but kept lean. The public roadmap adds scheduled audits, client reports and local SEO — a sign the project aims at real use in agencies and companies, not at the demo.

One item deserves attention because it's new compared to the classic suites: AI visibility, that is, understanding whether and how your site gets cited by AI engine answers. It's the ground of GEO (Generative Engine Optimization): with more and more searches ending inside an assistant, knowing if you show up in generated answers matters as much as ranking on Google.

• Keyword research and rank tracking for volumes, difficulty and positions over time.
• Competitor insights and backlinks to map who outranks you and why.
• Site audit for the on-page technical errors that hold back indexing.
• AI visibility to measure your presence in generative engine answers.

⚙️ How to use it: self-hosting and connecting the agent

For personal use the recommended path is Docker: it starts in single-user mode with no authentication, perfect locally. If instead you want it for a team or across devices, there's the deploy on Cloudflare Workers — free-plan compatible, with database backups and a SaaS-like experience. Search Console is optional and connects with your own Google OAuth client in about ten minutes.

Once the panel is up, you connect the agent from the AI & Agents section: OpenSEO hands you the MCP server configuration to paste into Claude Code or Codex, then you install the skills from the official docs. From that moment your project's keywords and backlinks are a context the agent can query whenever needed.

seo-project-setup     # set up the project and goals
seo-coach             # step-by-step guidance on priorities
keyword-research      # volumes, difficulty, intent
keyword-clustering    # group keywords by theme
competitive-landscape # overview of competitors
competitor-analysis   # focused comparison with one rival
link-prospecting      # find backlink opportunities

💰 What it really costs: the pay-as-you-go model

«Free» needs an honest footnote. The OpenSEO app costs zero because you self-host it; the data, though, comes from DataForSEO, an independent paid service. You pay only for the calls you make: a new account includes $1 of trial credit and the minimum top-up is $50, so you can test everything before deciding whether to open your wallet.

The orders of magnitude are low for anyone who isn't an enterprise agency: tracking 100 keywords weekly costs around $1.20 a month, a hundred keyword searches about $3.50, a hundred domain overviews about $4. The advantage isn't only price: it's predictability. No tier jumps or features locked behind the higher plan — you pay for the data, full stop.

repo      every-app/open-seo
stars     ~3,600 · MIT license
activity  latest release v0.0.21 (2026-06-24)
stack     TypeScript · Cloudflare Workers · Drizzle
data      DataForSEO (pay-as-you-go, min. $50)

⚖️ OpenSEO or the closed suites: the trade-offs

Let's not oversell it: Semrush and Ahrefs remain mature tools, with huge proprietary datasets, advanced features and support. OpenSEO doesn't replace them in every respect, and as a young project it should be validated on your own case before entrusting it with critical work.

The point is that for a great many scenarios — the SME tracking a few dozen keywords, the freelancer running the odd audit, the dev team that wants SEO data inside the agent — the self-hosted, on-demand model is simply a better fit. You control the data, you spend only what you use, and you get an MCP hook the closed suites don't offer today.

🏢 Why it matters for developers and SMEs

For developers, OpenSEO is a case study before it's a tool: a real TypeScript app on Cloudflare Workers that packages an MCP server and well-made agent skills. It's exactly the pattern we describe when explaining how to bring a business datum into an agentic flow — here you can read the code, fork it and adapt it.

For an SME the value is concrete: SEO stops being an opaque subscription and becomes a controllable process, where the data stays yours and the cost follows usage. And the agent hook changes the scale: instead of reading reports, you ask your assistant to analyze competitors or clean up a keyword cluster, with the real numbers underneath. It's the difference between «we have a SEO tool» and «SEO is part of how we work»

🗺️ Where to start

The path we suggest is low-risk: start OpenSEO locally with Docker, create a DataForSEO account and spend the trial credit on a real but non-critical project — your own site or a friendly client's. Run a keyword-research from the panel, then connect the MCP server to Claude Code and repeat the same search by asking the agent: that's where you find out whether the agentic flow really saves you time.

From there, look at the clustering and competitor-analysis skills and try AI visibility: it's the most current part, the one that tells you whether you exist in the answers assistants generate. Studying how OpenSEO stitches together data, MCP and skills is, in itself, an excellent lesson in agentic engineering — reusable well beyond SEO.

More articles from the blog