Talk to Your SEO Data — From Claude, Cursor, or Any LLM
The SignalSumo Model Context Protocol server lets any MCP-capable LLM client discover and call your SEO tools in natural language. Rank tracker, backlinks, GSC, AI visibility, site audit — 12 tools ready, OAuth 2.1 one-click connect, nothing to install.
No install — paste one URL into Claude or ChatGPT. Included on Pro & Agency plans.
What is the Model Context Protocol?
The Model Context Protocol (MCP) is an open standard from Anthropic that lets any LLM client — Claude, Cursor, Claude Code, and the growing set of MCP-capable IDEs — discover and call tools on a remote server. Instead of writing custom API code for every LLM you use, the model reads a plain-English catalog of tools, extracts arguments from your chat message, calls the right one, and quotes the result back. The SignalSumo MCP server exposes your rank tracker, backlinks, GSC, AI visibility and site audit as MCP tools — so a conversation like "which of my keywords lost the most rank this month" becomes a real, sourced answer instead of a promise.
Key Takeaways
- Twelve tools shipped today across rank tracking, backlinks, GSC, AI visibility, site audit and keyword research.
- Streamable HTTP transport, hosted and multi-tenant — works in Claude and ChatGPT with no install.
- Two auth modes: raw API key (local install) and full OAuth 2.1 with dynamic client registration + PKCE (hosted).
- Read-only tools (list keywords, GSC data, AI visibility) are free — they don\'t count against the monthly API cap.
- Session-isolated multi-tenant support — each connected user gets their own client instance, keys never cross.
- OAuth 2.1 with dynamic client registration and PKCE — no API key to copy or store.
Every question you ask the LLM about your SEO is a copy-paste job
The value of your SEO data is bounded by how easy it is to ask a question of it. Today, it isn\'t easy.
Screenshot & paste
You open the rank tracker, screenshot the losing keywords, paste it into Claude with a prompt. The image OCRs poorly, half the columns don\'t make it, and you re-type the numbers.
CSV round trips
Export GSC to CSV, upload to Claude, ask a question, upload backlinks CSV, ask again. Every question is a new export, a new upload, a stale snapshot.
Custom API code
The advanced way: wire the SignalSumo API into your own agent code. Works — but you rebuild it for every new LLM you try, and it\'s brittle when your team wants to use it too.
The answer is stale
By the time the CSV loads, the interesting question has moved. Live data hooked into the LLM is the only version of this that keeps up with the work.
Every SEO question is a one-shot tool call away
Five modules, twelve tools, one connection. Every tool calls the same SignalSumo API endpoint the web app uses, with the same authentication, plan limits and quota accounting.
Meta & usage
Rank tracker
Google Search Console
Backlinks
AI visibility
Actions (async jobs)
From "add MCP server" to "ask your data" in under a minute
Paste one URL into Claude or ChatGPT. No install, no API key, all 12 tools.
Add the connector
Claude: Settings → Connectors → Add custom connector.
ChatGPT: Settings → Plugins → Advanced → Developer mode on, then add a connector.
Paste https://signalsumo.com/mcp — nothing to install.
Authorize
Paste your API key once, or click "Authorize" in the OAuth consent browser tab. Access refreshes silently.
Discover
The client auto-fetches the 12 tools and their argument schemas. No setup UI, no manual mapping.
Chat
Ask a real question. The model picks the right tool, calls it, and quotes the answer with source data.
What talking to your SEO data actually looks like
Three real questions, three tool calls, three sourced answers. No copy-paste, no CSVs.
/rank-tracker in the same window?/rank-tracker — both dofollow, both from newsletters not comparison pages. The click drop isn\'t a backlink story; more likely on-page (title or CTR)./rank-tracker specifically: title tag length dropped to 38 chars in the last deploy — CTR-sensitive. Try lengthening it to 55-60 chars and re-check position in a week.All three questions answered from live data. Zero exports, zero screenshots.
MCP vs raw API vs "just paste it into Claude"
All three connect an LLM to your data. Only one keeps up when the questions get real.
| Feature | SignalSumo MCP | Raw API + custom code | Copy-paste to Claude |
|---|---|---|---|
| Setup time | ~60 seconds | Hours per integration | Zero — and it shows |
| Works across LLMs | Any MCP client | Rewrite per model | Yes, but same friction each time |
| Data freshness | Live on every question | Live if you wrote it that way | As stale as your export |
| Auth | OAuth 2.1 or API key | You build it | N/A — no server calls |
| Multi-turn conversation | Model chains tools itself | Yes with agent framework | Re-paste every turn |
| Cost per read | Free — DB reads uncapped | Same | Your time |
More API calls as you grow
Every paid plan includes the full toolkit. Plans differ only in how much you can use.
The tools your MCP conversation reads from
Rank Tracker
The keyword-rank data behind list_tracked_keywords + get_keyword_rank_history.
Rank tracker →Technical SEO Audit
The audit crawl start_site_audit kicks off, then get_job_status polls.
Website audit →How to fold MCP into your day without changing everything
MCP isn\'t a replacement for the dashboard — it\'s a shortcut for the questions the dashboard makes you click through five screens to answer. Here\'s the pattern that sticks.
Start with "what changed"
Every session opens the same way: "what moved on the site this week?" The model calls list_tracked_keywords + get_search_console_queries, cross-references, and gives you a two-sentence Monday briefing. Faster than the dashboard because it stops before the graphs.
Chain across modules
"Which page lost clicks AND which competitors gained backlinks in that window?" is one question to the model, two API calls under the hood. This is where the multi-turn thing pays for itself.
Fire actions, then walk away
Trigger an audit or keyword research from chat, get the job_id, then let the model poll for you in the background. Comes back with the finished result when you\'re ready — no browser tab, no reminder.
Keep the dashboard for the graphs
MCP wins on ad-hoc questions and cross-module reasoning. The dashboard still wins on graphs, snapshots, and everything you want to hand a client. Use both — that\'s the point.
Two ways to connect
The hosted endpoint is the current one: nothing to install, OAuth instead of an API key, and it carries the full set of 12 tools documented on this page. The npm package is an earlier local-install route that ships its own slightly different tool names — use the hosted endpoint unless you specifically need an offline install.
Claude
Settings → Connectors → Add custom connector. Paste the URL, sign in when the tab opens, click Authorize. Works in Claude on the web and in Claude Desktop, and the twelve tools appear in your next message.
ChatGPT
Settings → Plugins → Advanced settings → turn on Developer mode, then add a connector with the same URL. Available on Plus, Pro, Business, Enterprise and Education, on the web app. Same OAuth sign-in, same twelve tools.
Security posture — what the LLM can and can\'t do
The LLM inherits your API key\'s permissions — no more. Billing, plan changes, team management and admin endpoints are never exposed as tools.
Kill a key from the dashboard and the LLM loses access instantly. OAuth tokens live 1 hour; refresh tokens 30 days, rotated on every use per OAuth 2.1.
Every tool call goes through the same api_usage_log the REST API uses. You see exactly which tools the model called, when, from which key.