Flip through this short guide and you will have everything you need to get started: what to ask, how to connect it, and how it fits your workflow. Scroll past it for the full reference and FAQ.
Six quick slidesFive-minute readNo sign-up to read
What you can ask
A dozen things your agent can now answer.
Each tool is one question, answered with live data. Ask in plain words; the agent picks the tool.
Research
1 lookup each
keyword_researchRelated keywords with volume, difficulty and intent.
keyword_difficultyHow hard one keyword is to rank for.
serp_overviewWho ranks in Google's top 10.
domain_overviewA domain's traffic, keyword count and authority.
ranked_keywordsWhat a domain already ranks for.
competitor_keywordsThe domains chasing your keywords.
ai_keyword_volumeHow often a term is asked in AI, not Google.
Backlinks
3 lookups each
backlink_overviewAuthority, total backlinks and spam score.
referring_domainsEvery site linking to a domain, ranked.
backlink_gapLinks your rivals have and you don't.
AI visibility
8 lookups each
ai_visibilityHow often ChatGPT and Google AI mention you.
ai_share_of_voiceYour share of AI answers vs competitors.
Your own data
free, 0 lookups
Google Search ConsoleYour real clicks, positions and queries.
check_creditsHow many lookups you have left.
Setup
Connect it once. Under a minute.
Your private MCP URL arrives by email after you subscribe. Add it to your agent and you are ready to go.
01
Claude Code
One command in your terminal:
claude mcp add --transport http contextbolt-seo <your URL>
02
Claude Desktop · Claude.ai
Settings, then Connectors, then Add custom connector, give it any name, and paste your URL. On Claude.ai in the browser, tool calls may show a connector ID instead of the name. That is normal.
03
Cursor
Add to your mcp.json under mcpServers:
"contextbolt-seo": { "url": "<your URL>" }
04
Codex
Add to your ~/.codex/config.toml:
[mcp_servers.contextbolt-seo] url = "<your URL>"
First session
See it work, then ask your own.
Not sure it connected? Ask what SEO tools do you have? If it lists keyword and SERP tools, you are good. Watch it run below, then point it at your own site.
Claude Desktop live data
Click to try a real question
In your workflow
How it fits your SEO process.
Real jobs, start to finish. Ask the whole question and the agent chains the tools for you.
Plan a new post
Find a keyword you can rank for, then see who you are up against.
keyword_research→keyword_difficulty→serp_overview
Audit a competitor
Size up a domain, see what it ranks for, find who else is in the race.
Find who links to your rivals but not you, then judge each prospect.
backlink_gap→backlink_overview
See if AI recommends you
Check whether AI answers mention you, then see who they name instead.
ai_visibility→ai_share_of_voice
Get the most out of it
Small habits, better answers.
Ask in plain English. No filters or query builders. "Who ranks for X?" just works.
Let it chain tools. Ask one big question and the agent calls several tools to answer it.
Be specific. Name the exact keyword or domain and the answer gets sharper.
It remembers. Ask about a keyword again and the answer leads with what has changed since last time. Memory and project files are automatic and never count against your 1,000.
Findings live in your project. In Claude Code, Cursor or Codex, every lookup writes to ./seo-findings/ as markdown. Grep it, commit it, drop it in Obsidian.
Most calls are one lookup. The $35 plan includes 1,000 a month, far more than most people use. Research tools cost one lookup each, backlink tools three, and the two AI-visibility tools eight. Your own Google Search Console is free.
Checking your balance is free. Ask how many credits you have left anytime, even at zero.
Trust the direction. Numbers are decision-useful estimates, not identical to Ahrefs.
Say you are launching an AI note-taker app and want to find the SEO opening. Drive the demo below, then see where the thread goes next.
Claude Desktop live data
Click to try a real question
The demo is the first few questions. You keep pulling the thread, and each follow-up is one more plain question that tells you what to do next.
"What does otter.ai already rank for?"ranked_keywords
Their biggest traffic pages, which double as a ready-made list of topics you can cover too.
"Who else competes for these keywords?"competitor_keywords
The full set of rivals, including a couple you would never have named off the top of your head.
Ten minutes, no dashboard, no exports. You went from a blank page to a keyword shortlist, a read on the market leader, and a content plan, in the chat you already work in.
A week later, the memory pays off. Ask "how is 'meeting summary generator' looking now?" and your agent's answer leads with what has changed: "Since last lookup 7 days ago, difficulty has gone from 29 to 32 and search volume from 1,900 to 2,100." No need to remember the old numbers, no extra credit spent on memory. And if you ran this in Claude Code or Cursor, the whole session is already on disk in ./seo-findings/ as markdown you can search, commit, or open in Obsidian.
Backlinks, too. Ask "who links to otter.ai but not me?" and backlink_gap hands back a spam-scored prospect list, a ready-made outreach shortlist. You also get backlink_overview for a domain's authority and referring_domains for who links to it. Backlink calls cost three lookups each, since that data is more expensive to pull.
Keep learning
Go a little deeper.
Want the mechanics, or to see how ContextBolt SEO stacks up against the alternatives?
Ask your agent "what SEO tools do you have?" If it lists keyword, SERP, domain and competitor tools, you are connected. If not, re-check the URL you pasted.
Can I ask follow-up questions?
Yes. It is a normal chat. Ask "now show me who ranks for it" or "which of these is easiest?" and the agent keeps the thread, calling more tools as needed.
Does it know which site is mine?
Tell it once in the chat, like "my site is yourblog.com", and it uses that for the rest of the conversation. There is no account setting to configure.
Does it remember what I have researched before?
Yes. Every lookup is kept on your account, and when you ask about the same keyword or domain again your agent's answer starts with what has changed since last time. You can also ask "what have I researched?" or tell the agent to pin a finding. Memory is free and does not draw from your monthly lookups.
What is the seo-findings folder?
When you connect ContextBolt SEO in Claude Code, Cursor or another client with file access, every lookup also saves to a ./seo-findings/ folder in your project as markdown. One file per keyword or domain you research, plus an INDEX. You can search it, commit it to git, or open it in Obsidian. claude.ai web users still get the same recalled research; the files just do not appear on disk.
What happens if I hit the limit mid-task?
The tool returns a friendly message with your reset date and a top-up link, and makes no charge. Nothing breaks. Add a top-up pack and carry on.
Can I use it for client or freelance work?
Yes. Run research for any domain, yours or a client's. One subscription, no per-seat fees.
Do I need to know SEO jargon?
No. Ask in everyday language. If a term like difficulty or CPC comes back, ask the agent to explain it.