Connect your coding agent
Give Claude Code, Cursor or any MCP-capable agent your site’s fix list, so it can make the changes, check them on the live page and close them out.
SEOgent works out what your site needs. Your coding agent makes the changes. The two talk over MCP, so the agent working in your site’s repository can read your weekly briefing, pick up fixes that are ready, check the live page once you deploy, and mark the work done.
1. Create a token
Tokens are made by an account admin under Settings → Agent access.
- Give the token a name that says where it’s used, such as “Claude Code on my laptop”.
- Choose the site it’s for.
- Click Create token.
SEOgent shows the token once, along with ready-to-paste setup for Claude Code and Cursor. Copy it before you leave the page. If you lose it, revoke it and make a new one.
Each token sees one site only. If you look after several sites, make a token for each and set each one up in that site’s repository. A token keeps working until you revoke it, and stops while your plan is inactive.
2. Add SEOgent to your agent
The server address is:
https://app.seogent.ai/mcp/seoIt uses MCP’s Streamable HTTP transport, with your token as a bearer token in the Authorization header.
Claude Code
Run this in your site’s repository, with your token in place of YOUR_TOKEN:
claude mcp add --transport http seogent https://app.seogent.ai/mcp/seo \
--header "Authorization: Bearer YOUR_TOKEN"Run /mcp inside Claude Code to check that seogent is connected.
Cursor
Add this to .cursor/mcp.json in your site’s repository:
{
"mcpServers": {
"seogent": {
"url": "https://app.seogent.ai/mcp/seo",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}That file now holds your token, so add .cursor/mcp.json to your .gitignore before you commit.
Other agents
Any agent that can connect to a remote MCP server over HTTP works. Give it the server address above and send the header Authorization: Bearer YOUR_TOKEN.
3. Ask your agent to start
You don’t need to explain the tools. SEOgent tells your agent how the session works when it connects. A prompt like this is enough:
Check SEOgent for SEO work that’s ready on this site, and start with the highest-impact item.
A typical session goes like this:
- Read the briefing to see what changed this week and what matters most.
- List the ready work. These are action items with a fix plan, and approved articles waiting to be published, highest impact first.
- Get the instructions for one item: what to change, where, the exact new values, and how to check it worked.
- Make the change in your code, following how the site is already built. Titles, meta tags, canonicals, headings and structured data go in the server-rendered HTML, because AI crawlers don’t run JavaScript.
- Check the live page once you’ve deployed, to confirm the change is there.
- Close the item with a short note of what changed. The next daily crawl checks the page again.
Your agent works on one item at a time and only changes what the task asks. Some fixes include steps for you, such as changes in Search Console or your hosting dashboard. Those are listed separately, and your agent leaves them for you.
What your agent can do
| Tool | What it does |
|---|---|
get_briefing | Reads the latest weekly briefing: the headline, summary, wins and concerns |
list_ready_work | Lists action items that have a fix plan and approved articles waiting to be published, highest impact first |
list_open_issues | Lists open items that don’t have a fix plan yet and what each is waiting on. For context only; your agent won’t change code for these |
get_task | Gets the full instructions for one action item, including the fix plan’s steps and how to check the result |
get_content | Gets an approved article as a Markdown file with frontmatter, plus the address it should be published at |
check_page | Loads a live page the way search engines and AI crawlers see it, without JavaScript, and scores it for SEO and AEO |
get_ai_visibility | Shows how often ChatGPT, Claude, Gemini and Perplexity name or cite the site, which competitors they name instead, and the questions competitors win |
complete_task | Marks an action item done, with a note of what changed |
mark_published | Records that an approved article is live, so its search results are tracked from that day |
list_sites | Lists the site the token can reach and how many items it has open |
Only complete_task and mark_published change anything in SEOgent. Everything else just reads.
Only planned work reaches your agent
Your agent only works on items you’ve planned in SEOgent. New issues show up in list_open_issues with what each is waiting on, but they don’t reach list_ready_work until you plan a fix from the item in SEOgent, or approve the article in your content plan. You decide what gets changed, and your agent does the work.
Revoking a token
Revoke tokens you no longer use, or any that may have been shared by mistake, under Settings → Agent access → Tokens. Each row shows when the token was last used. Agents using a revoked token lose access straight away.
Troubleshooting
- The agent says it isn’t authorized. Check that the header is exactly
Authorization: Bearerfollowed by your token, and that the token hasn’t been revoked. - “SEOgent’s plan for this account isn’t active.” An account admin can restart the plan under Settings → Billing.
- “This token only has access to example.com.” The token is for a different site. Create one for this site and use it in this site’s repository.
- There’s no ready work. Every open item is still waiting for a fix plan. Ask your agent to run
list_open_issuesto see what each one needs, then plan the ones you want done in SEOgent. check_pagedoesn’t see the change. It loads the live page, so deploy first. If you use a CDN, the old page may still be cached.
Updated October 27, 2026 Next: How fixes work →