GitHub MCP for SEO
Connect SEOmatic's MCP server and your AI agent finds what is holding your GitHub HTML pages and Markdown posts back in Google, stages the fix through pull requests on your repository, and waits for your approval. Then it checks whether the change worked.
- Your real Search Console data
- Fixes staged in GitHub
- Your approval first, by default

Last Checked
October 2026
Endpoint
https://app.seomatic.ai/api/mcp
Full Docs
Every tool, parameter and cost: SEOmatic MCP docs.
How SEOmatic's MCP Works on GitHub
The Short Answer
GitHub's official MCP server lets an agent work with repositories, issues, pull requests and Actions. SEOmatic's MCP server (https://app.seomatic.ai/api/mcp) does a different job on GitHub: SEO. It reads your Google Search Console data, finds the HTML pages and Markdown posts that are losing clicks or not indexed, and stages the fix through pull requests on your repository. By default nothing goes live until you approve it, every change can be rolled back, and each one is measured afterwards and marked verified, ineffective or regressed. You can run both servers in the same agent.
Which GitHub MCP Do You Need?
| SEOmatic MCP | GitHub MCP (official) | |
|---|---|---|
| What it is for | SEO: find what costs clicks, fix it, measure it | Work with repos, issues, pull requests and Actions |
| Your Search Console data | Yes, free with no limit | No |
| Picks what to change from search data | Yes | No |
| Rollback | Yes, with a check for edits made since | Not built in |
| Measures whether a change worked | Yes | No |
What SEOmatic Can Change on GitHub
A repository has no fields and no API for SEO, only files. So the agent has to earn the right to edit one: it maps each live URL to the file that produces it, checks that the file's title matches the page, and opens a pull request only when the match holds up (a page with no title to compare is matched on its path alone). When it is not, you get a note instead of a guess.
- HTML Pages, Fully: On plain HTML files, agents set the title, meta description, canonical tag and structured data, and rewrite content inside your page's existing main element. Structured data sits in a clearly marked block, so your own markup is never touched.
- Markdown and MDX Front Matter: On Markdown and MDX files, agents update the title and description in the front matter. The body is left alone, because converting it through HTML and back breaks its formatting and can break an MDX build, and a lossy edit is worse than none.
- New Blog Posts in Batches: On a Markdown blog, programmatic pages arrive as one pull request per batch, all in a single commit with a manifest. The next batch is queued only once you merge the last one, and undoing a batch is reverting one commit.
- Next.js Pages Are Read, Not Rewritten: React page files are read so agents understand your site and can diagnose what is holding a page back, but they are never edited. Having an AI rewrite JSX is a risk we decline, and the fix comes to you as a written change instead.
- Your Main Branch Is Off Limits: The connector refuses to commit to your default branch. Every change is a branch and a pull request, so it goes through the same review, checks and preview builds as your team's own code.
- Applied Means Live: A merged pull request still has to build and deploy. The change counts as applied only once your live page shows it. If bot protection blocks that check, it confirms the change in the merged files instead and labels it that way. An open pull request never counts, and if the deploy never lands, it says so rather than reporting a success.
Every field, limit and safeguard is on the GitHub integration page.
Connect GitHub and Your Agent
1. Connect GitHub to SEOmatic. In SEOmatic, go to Connections, select GitHub, and click Continue with GitHub. On GitHub, install the app on the one repository your site lives in. GitHub sends you back to SEOmatic.
2. Connect SEOmatic to your agent. Add https://app.seomatic.ai/api/mcp as a remote MCP server (Streamable HTTP) in your agent, then sign in to SEOmatic in the browser. Sign-in is OAuth 2.1 with PKCE and dynamic client registration, so there is no key to paste. Scripts and headless agents can send a workspace API key as a Bearer token instead.
https://app.seomatic.ai/api/mcpWhat to Ask About Your GitHub Site
- "Which pages built from my repository lose the most clicks?"
- "Open a pull request with better titles for my Markdown posts with low CTR."
- "Find pages Google has not indexed and tell me why."
- "Which posts rank 5 to 15 and could move up with a refresh?"
- "Audit my live site and point me to the files behind each issue."
Fix GitHub SEO From Your Agent
Setup guides for GitHub with each agent: Claude, Claude Code, Claude Cowork, ChatGPT, Codex, Cursor, Gemini, OpenClaw, Hermes Agent, Grok Bot, Perplexity Computer, Manus. Setup guides per agent: Claude, ChatGPT, Claude Code, Cursor, Codex, Gemini and 24 more.
From Answer to Fix
By default, nothing changes on your site until you approve it. A fix is staged as a proposal that says what will change and why. Approved fixes run within seconds, the page's previous state is stored, and any edit to an existing page can be rolled back. A rollback refuses to overwrite edits someone made since. Changes that could take a page out of Google, like a noindex, need an extra confirmation.
Every shipped change is measured against its own baseline and the site's trend: after 14 days for a title, 28 for a content refresh, 42 for striking-distance and cannibalization work, 90 for a new page. It is marked verified only when the gain over the site's own trend is statistically clear and worth at least 5 more clicks; otherwise it is ineffective, or regressed when it falls more than 15%. Low-risk changes that lose traffic roll back on their own.
What Is Free
Connecting is free with no card. Search Console reports, page checks and 30-page site audits are free, as often as you like. Live market data (keyword volumes, competitors, backlinks) uses free questions: 5 a month, plus a one-time 10 when you connect Search Console, and one question covers up to 5 lookups. Staging and publishing fixes is part of paid plans, which start with a 14-day trial.
More SEOmatic MCP Pages
Frequently Asked Questions
Is there an official GitHub MCP server?
Yes. GitHub's official MCP server lets an agent work with repositories, issues, pull requests and Actions.
How is SEOmatic's MCP different on GitHub?
It is built for SEO: it reads your Search Console data, stages fixes to your GitHub HTML pages and Markdown posts for your approval, and measures whether each change worked. GitHub's own MCP options are used to work with repos, issues, pull requests and Actions.
What do I need to connect GitHub?
A site whose pages live in a GitHub repository as HTML or Markdown files, and permission to install a GitHub App on that repository. The app asks for access to contents and pull requests on the one repository you choose.
Will SEOmatic publish to GitHub without asking?
No. By default every change waits for your approval, your agent is told never to approve on its own, and any applied change can be rolled back.


