DataForSEO MCP in Claude Code: what it costs

A monthly SEO suite charges its plan fee whether you query it or not; DataForSEO data called from Claude Code is billed per request. We tested both versions of the server in one half-hour session on 26 August 2026: it cost $2.35, across eight of DataForSEO’s paid APIs, and the account’s balance and its day-spend counter agree on the exact total, $2.346159, to the millionth of a dollar. This page shows that DataForSEO bill, how we checked it, and what we still run through it; any Claude Code subscription or model cost is separate and not counted here. The session ran against wiserstaff.com, one of the sites we manage.

What the DataForSEO MCP server is

The DataForSEO MCP server is a free connector that lets an AI assistant call DataForSEO’s paid search-data APIs in conversation. You ask Claude Code a question in plain language; the assistant picks the request, sends it with your API login, and reads the answer back. The server costs nothing. Every paid request it sends is billed to your DataForSEO balance.

MCP stands for Model Context Protocol, the standard Claude Code uses to plug in outside tools. DataForSEO publishes its server’s code under the Apache-2.0 licence. Its repository, read on 17 October 2026, describes a current version built on four tools, three that read the documentation and one, api_request, that can reach any endpoint, and marks the earlier multi-tool version as deprecated. Our session ran both: the earlier one with a typed tool for each endpoint, the current one beside it.

If DataForSEO itself is new to you, start with what DataForSEO is and how it bills: it covers the APIs, the $50 minimum payment and the habits that keep the spending under control. This page assumes you know the service and want to know what running it from Claude Code costs; the installation steps are on DataForSEO’s own MCP page.

What one half-hour session cost us

One half-hour session through both versions of the server cost $2.35 on 26 August 2026, across eight paid APIs: SERP, Keywords Data, DataForSEO Labs, Backlinks, OnPage, Domain Analytics, Content Analysis and AI Optimization. Of the itemised APIs, Backlinks and Labs cost most; the part the account does not itemise cost more than either. All came from our prepaid balance.

The account keeps a cost record for each task in seven of those APIs. AI Optimization has no such record, and neither has DataForSEO Trends, the vendor’s own trend service, which belongs to Keywords Data but is left out of its record; we sent it three requests. Their combined share is the balance drop minus the itemised total.

The seven itemised lines add up to $1.3651, 58% of the total. Figures are rounded to four decimals so that they add up, which is why the total shows $2.3462 rather than the exact $2.346159; the session has its own dated line in the cost rows of our figures page.

Most of the requests the account itemised cost between $0.002 and $0.025 each. The dearest were two Google Ads volume requests at $0.09 each, about seven times a typical Labs call at about $0.013, and two domain-level requests at about $0.12 each, about nine times: a domain history in Labs and a domain record in Domain Analytics. Knowing the prices of the requests you plan helps, but the total still depends on how many the assistant makes.

The two-counter check: how we reconcile the bill

The two-counter check confirms a session’s cost from two separate numbers the account keeps: the balance, read before and after, and the day’s spending counter, read at the same moments. If both give the same total, the figure is the account’s own, not a sum added up from responses. On 26 August 2026 both said $2.346159.

We read both from the account’s user_data endpoint, which is free to call:

We ran nothing else on the account during that half-hour, so both counters measure the session alone, and their agreement to the sixth decimal place means the session cost what the account says, whatever the responses showed on the way. That matters with an assistant in the loop: the earlier server version leaves the cost field out of most of its answers, so a total built from the replies would have understated the spend. We put the rule in our house rules: a number without a source does not exist, and the source for spending is the account.

The check needs nothing but two free reads. What it does not tell you is which request cost what; for the seven APIs that keep per-task records, the account gives that too.

What we run through it now

Since the test session, the MCP server answers the one-off questions that come up while we write or audit: a keyword overview before an article, a volume check, a quick look at a competitor’s links. Repeated jobs, such as reading the top Google results before each article, run from scripts against the same account and the same balance.

Some reference points from that account, MCP and scripts together, all dated:

• One article bench: three Google result pages read live, depth 10, by script, $0.006 (15 October 2026, the bench for this article, separate from the test session).
• A 54-topic blog plan: keyword research for $0.41, under a cap of $1.00 written before the first call (5 September 2026, on our figures page).
• A full month of itemised requests: $1.85 in September 2026 for every site we manage, AI Optimization and DataForSEO Trends not included. Itemised spend was highest in August, the month of the test session.
• Everything since our deposit: $9.68 from the $50 deposit of 10 August 2026 to a balance of $40.32 read on 17 October 2026, which is 68 days, with no other deposit in between. That figure is the deposit minus the balance, so it counts every request on the account, MCP and scripts together, AI answers included.

For a solo operator weighing this against a subscription, the comparison with a suite’s monthly plan is in the tools we use instead of Semrush. For how automated jobs fit around a manual review, see the SEO tasks we automate.

MCP in Claude Code or a script?

We use the MCP server in Claude Code when one person needs one answer now, and a script when a job repeats, runs in bulk or must stay inside a written budget. The MCP call is quick to ask and easy to read; our scripts stop at a written cap, log every request and record the balance before and after.

The split follows from what each does best:

• In conversation, the assistant chooses the request, reads the reply and explains it. That suits a keyword overview before writing, a single volume check or a question about an endpoint you have not used.
• In a script, the same request runs a hundred times without a person watching, so the spending rule has to be written into the code, not remembered.

Your account publishes its own price list alongside the balance, so both can be read before any paid call; we describe that habit in our By-the-Call SEO Manual. The principle fits in one line: decide what a session may spend before the assistant starts, and read the account when it ends.

Who should not use it

The DataForSEO MCP server is a poor fit for anyone who wants a dashboard, a saved project and charts without asking for them. It returns raw data to an assistant, and someone has to know what to ask. It also suits nobody who will not look at a prepaid balance: the bill depends entirely on the requests made.

If you only need familiar reports on your own site’s search performance, a free console such as Google Search Console may be enough; if you want those reports ready-made for competitors too, a subscription suite will serve you better than an assistant that needs instructions. If you want to try raw data without a deposit first, DataForSEO’s documentation describes a free sandbox that returns sample responses.

What we did not verify

The half-hour test ran locally in Claude Code; the later reference points mix that connector and our scripts, as each one says. We did not run DataForSEO’s hosted remote endpoint, which bills the same account, nor the server in Cursor, ChatGPT or Gemini CLI.

The repository and MCP-page details are as read on 17 October 2026. Prices are those of our own account on the dates given; yours may differ, so read your account’s price list before you plan a budget.

What the DataForSEO MCP costs in Claude Code, then, is whatever the requests cost: $2.35 for our half-hour test across eight APIs, within which one live page of Google results cost $0.002. Decide on a spending limit before the assistant starts, note both counters, and read them again when it ends: the day counter does not start at zero.