The open source SEO tool stack for solo devs

SEO tools are priced for an agency running twenty client sites. I run three. That gap is the whole reason my setup is an API key and a folder of agent commands instead of a subscription — the same logic that keeps the rest of my build-in-public stack small.

Why the subscription is the wrong shape

Not because it's expensive — because it's a seat, billed monthly, whether or not you use it. Solo SEO work happens in bursts: a hard week before a launch, then nothing for a month. You pay for the quiet months at the same rate.

The second problem is what you get for the money. Here's a real keyword card from a research session: seo services, difficulty Easy, volume >10,000. Both numbers are true and the conclusion they imply is wrong. Look at the live results and the first three organic spots are Fiverr and two agencies, under a local pack of three more agencies. It's a hire-someone query. Nothing you write ranks there. The monthly volume series also runs 165K down to 27K over the past year — the number on the card is a trailing average of a market that's collapsing into AI Overviews.

A keyword tool sells you a number. The decision is a separate job, and no plan tier includes it.

The three layers

Data: DataForSEO. Pay-as-you-go per API call — SERPs, keyword metrics, backlinks, on-page. Every response includes the exact cost of that call, so spend is observable instead of estimated. No seat, no minimum.

Interface: OpenSEO. Open source (MIT), and explicitly an alternative to Ahrefs and Semrush: keyword research, rank tracking, competitor insights, backlinks, site audits, AI visibility. Self-host with Docker locally, or on Cloudflare's free tier if you want it reachable by a team. You bring your own DataForSEO key, so the tool is free and you pay the API directly. There's a hosted version at $10/mo that adds ~28% to the raw API cost if you'd rather not run it.

Judgment: MCP. This is the layer I care most about, and it's the one OpenSEO gets right — it ships an MCP server plus reusable agent skills, so Claude Code can pull the data itself rather than you copying numbers out of a dashboard.

Where the agent actually helps

Not at strategy. An agent is not better at SEO than you are, and prompting it for "SEO advice" returns the same blog-slop you'd have found anyway.

It's better at the part that's tedious enough that you skip it. Checking a keyword properly means pulling volume, difficulty, the live results, the trend series, and your own Search Console data, then reconciling five sources that disagree. That's four minutes of API calls and twenty minutes of cross-referencing, per keyword. An agent does it without getting bored on the eleventh one — which is the only reason it gets done at all.

Two things made mine useful rather than decorative. A knowledge base: one file of how search actually works, so it reasons from mechanics instead of listicles. A committed ledger: every action written down with a verify-by date, so a claim made in March gets checked in June. Without memory, an agent re-derives the same shallow conclusion every session.

The honest caveat

I haven't run OpenSEO in production. My own setup talks to DataForSEO and Search Console over MCP with no UI at all, because I wanted the agent to be the interface — the numbers that matter end up on my public page through its Search Console widget rather than in a dashboard I log into. So take this as an architecture recommendation, not a review. What I can vouch for is the shape: own the data source, rent nothing per-seat, and put the agent where the tedium is.

If you want that shape with a real interface on top instead of building it, OpenSEO is the one I'd try first. It's the same bet — data you pay for by the call, and an MCP server so the agent can drive it.