The RankForge agent, inside your editor
RankForge ships as an MCP server. Connect it to Claude Code, Cursor, VS Code, Windsurf or any MCP-capable assistant: the assistant measures your site with RankForge's deterministic engine, then fixes the code right in the repository it already has open.
Why this beats a dashboard
An audit report tells you what's wrong; your coding assistant can actually change the code. With RankForge connected over MCP, one prompt — “audit my dev server and fix what you find” — runs the full technical-SEO audit against localhost, walks the findings, patches your files, and re-audits to verify. The measurements are deterministic (no AI guessing); the reasoning is your assistant's.
Connect from Claude Code
Run RankForge locally (or use your deployed instance), then:
claude mcp add --transport http rankforge http://localhost:3000/api/mcpThat's it. Ask Claude Code: “Use rankforge to audit http://localhost:3000 and fix the issues you can.”
Connect from Cursor / VS Code / Windsurf
Add an HTTP MCP server to your client's configuration (e.g. .cursor/mcp.jsonor VS Code's mcp.json):
{
"mcpServers": {
"rankforge": {
"url": "http://localhost:3000/api/mcp"
}
}
}What the agent can do
- audit_site — full deterministic audit: category scores, every issue with evidence, orphan pages, click depth, duplicate-content clusters, semantic internal-link suggestions, measured image weights and broken links. Works on localhost.
- audit_page — instant single-page check while you iterate.
- get_fix_template — framework-idiomatic patch (unified diff) for mechanical issues: robots, sitemap, viewport, lang, canonical, structured data, OpenGraph image.
- seo_docs — grounded answers from the RankForge knowledge base, so the assistant explains issues accurately.
- list_rules — the full rule catalog with documentation links.
Local vs hosted
Running RankForge locally, the agent audits localhost — see issues as you build, before deploying. A hosted instance (e.g. on Vercel) audits public URLsinstead: it crawls from the server, so it can't reach your machine's localhost. Point it at your live or staging site; to audit a dev server, run RankForge on your own machine.
Security model
In development the endpoint is open on localhost. A deployed instance is locked by default and refuses to probe private or internal addresses — it validates every resolved IP, including through redirects. To let people connect to a hosted instance, set RANKFORGE_MCP_PUBLIC=1 to open it, or RANKFORGE_MCP_KEY to require a Bearer token. Either way the tools only read rendered pages — the same thing any crawler sees — and nothing touches your repository without a human merging a pull request.