# Claude Code agent for SEO blogs, without burning limits or credits > Source: > Published: 2026-11-08 15:59:22+00:00 A Claude Code agent for SEO blogs can research topics, write drafts and check the finished pages, but every step spends from two meters at once: your Claude plan's usage limits and the credits of whatever SEO tool the agent calls. Run it carelessly and you hit a usage limit halfway through a draft or pay for the same keyword lookup three times. Here is a way to run it that keeps both in check. ## The two meters **Claude Code usage.** Long sessions that read many pages and write long files use up your plan's session and weekly limits. Anthropic changes these limits from time to time, so check your current numbers in Claude Code's `/usage` view or a tracker rather than trusting an old blog post. **SEO tool credits.** With seogeoaeo.ai connected over MCP, each tool run is charged in credits. Keyword Research costs 34 credits a run, Keyword Metrics 33, Keyword Ideas 15, and the Deduplicator and Clusterer 3 each, as of November 2026 per the [tool catalog](https://seogeoaeo.ai/tools). Checking your balance is free. ## Split the work into three sessions Doing research, writing and checking in one long session is the most expensive way to run it. Split it instead: 1. **Research session.** The agent reads your product pages, lists seed topics, runs keyword research on them and saves every result to a file such as`research/keywords.json` . It ends with a table of chosen topics and focus keywords. 2. **Writing sessions.** One fresh session per post, started from the saved table. The agent reads the research file instead of calling the SEO tools again, so writing costs no credits at all. 3. **Check session.** After the posts are live, one session runs the On-Page SEO Checker (8 credits) on each URL with its focus keyword. Fresh sessions also keep the context small, which matters for usage: a writing session that does not carry forty pages of research text spends less. ## Cheap tools first Order the research calls from cheapest to most expensive: - Merge duplicates with the Keyword Deduplicator before scoring anything, so you never pay to score "meta ad library" and "meta ads library" separately. - Score the whole candidate list in one Keyword Metrics run of up to 100 phrases rather than many single lookups. - Save Keyword Research for the seeds that survive, since each run covers only one seed. ## Put a cap in the prompt Tell the agent the most it may spend before asking, for example: "Before you spend more than 200 credits, show me the total and wait." The seogeoaeo.ai skill already tells the agent to confirm the credit total before a large batch, and scripts calling the REST API can pass `maxCredits` so a run above the limit is refused instead of charged. ## Do not pay twice Agents retry. If a tool call times out and the agent calls it again, a script can send the same `idempotencyKey` so the original run comes back instead of a new charge. In chat, the simpler rule is "read `research/` before running any tool". ## Watch the usage side too [UsageAtlas](https://usageatlas.com) is a free, local-first desktop tracker for AI coding tools, including Claude Code. It shows tokens, cost estimates, limits and reset times, so you can see before a writing session whether there is room to finish it. Its cost figures are estimates, not a bill. ## What a good run looks like One research session, a reviewed topic table, one short session per post and a final check pass. The SEO data comes from [seogeoaeo.ai](https://seogeoaeo.ai), whose plans start at $29 a month, and the agent's own usage stays visible the whole way. If you would rather not connect a paid tool, you can still split sessions and cache research by hand; you just lose the search numbers that make the topic choice worth trusting.