Grok supports custom MCP connectors, so it reads your Search Console data through the same endpoint every other assistant uses. Add mcp.gscwizard.com at grok.com/connectors, sign in once, and Grok works from real clicks, impressions, position and CTR instead of inventing them.
Endpoint: https://mcp.gscwizard.com/mcp
The short answer
Three steps, about two minutes:
https://mcp.gscwizard.com/mcp and complete the sign-in.No client ID, no secret, no API key: the server speaks OAuth 2.1 with dynamic client registration, so Grok registers itself and simply asks you to approve the connection. It then discovers the tools and offers them in conversations, exactly like a built-in connector.
A walkthrough of the same MCP server Grok connects to: adding it, authenticating, and asking for real Search Console analysis in plain language.
A custom MCP connector exposes any external API or tool to Grok, on every plan including the trial. GSC Wizard is a hosted MCP server, so there is nothing to install or run: point Grok at the URL and sign in.
Sign in at tool.gscwizard.com and connect the Search Console properties you want Grok to read. The MCP only ever sees properties already connected to your account.
Go to grok.com/connectors, click New Connector, then select Custom and enter the MCP server URL.
Approve the GSC Wizard sign-in and consent screen. Grok discovers the tools the server exposes and offers them in conversations, just like the built-in and catalog connectors.
MCP server URL
https://mcp.gscwizard.com/mcpGrok reaches the server from its own infrastructure, so the URL has to be public. That is what tunnelling services are for when an MCP server runs on your laptop - here there is nothing to tunnel, because GSC Wizard is hosted.
If your connector form offers a custom header instead of a sign-in, the server also accepts a static key: create one under Account → API keys and send Authorization: Bearer gscw_live_.... Keys carry a read-only or read-and-write scope, are shown once, and can be revoked at any time.
A connector gives Grok the data. A Bot gives it the method. A Grok Bot is a persistent agent with a fixed job, its own instructions and its own routines, so the rules that make SEO analysis trustworthy stay in place instead of being restated every session.
Instructions worth pasting in
Use GSC Wizard as the authoritative source for Search Console performance.
1. Use GSC Wizard tools instead of estimating data from web search.
2. If the property is unclear, start with list_sites.
3. Prefer the specialized server-side analysis tools over pulling large
raw Search Console datasets.
4. Always state the analysis and comparison periods.
5. Separate measured facts, interpretation, and recommendations.
6. Do not claim causation when the data only shows correlation.
7. Prioritize findings by likely organic-search impact.
8. Treat a lower numerical average position as better.
9. Do not do large arithmetic over raw rows when a deterministic
analysis tool already answers the question.
10. Never call a write tool without explicit user intent.Most of these exist because assistants get SEO data subtly wrong in predictable ways: averaging an average position, reading a lower position as worse, calling a seasonal dip a ranking loss, or quietly truncating a dataset and reporting the total anyway. Naming the failure modes up front is what turns a chat into a report you can send on.
Add a routine only once the same analysis has proven itself interactively. A scheduled report that quietly says the wrong thing is worse than no report.
Schedule around the data, not the calendar. Search Console settles on a two-to-three day lag, so a weekly report run on Monday for “last week” quietly drops Saturday and Sunday and shows a decline every single week. Run weekly jobs mid-week and monthly jobs a few days into the month.
Bots can be shared by link, which hands someone a copy of the configuration — never your credentials or your conversation history. They connect their own GSC Wizard account and see only their own properties.
Grok Build is xAI's terminal coding agent, and it supports remote MCP servers. Registering GSC Wizard there means the agent can read the performance of the pages it is about to change — which queries a template ranks for before you touch it, what a redirect did to a section, whether the page you are refactoring earns anything at all.
Register the server
{
"mcpServers": {
"gsc-wizard": {
"type": "http",
"url": "https://mcp.gscwizard.com/mcp"
}
}
}Grok Build runs a browser sign-in flow for MCP servers that use OAuth, so the key never has to touch the config file. If you would rather use a static header, keep the key in an environment variable and out of source control. Check the syntax against your installed version — Grok Build is moving quickly.
Grok Build plugins bundle an MCP server together with skills, slash commands and agents, so a whole analysis runs from one command instead of a paragraph of instructions. The GSC Wizard set covers the analyses people repeat most:
/gsc-reportWeekly review, week over week/gsc-rankingsWinners and losers by impact/gsc-decayRefresh candidates, diagnosed/gsc-cannibalizationCompeting pages, triaged/gsc-opportunitiesNear-term wins ranked by estimated click upsideBuilding an agent rather than chatting? Run the MCP client yourself, hand the tool list to a Grok model over the xAI API, and execute the calls it asks for. The xAI API is OpenAI-compatible, so the usual client works with a different base URL.
Python: MCP tools, Grok model
import json, os
from openai import OpenAI
from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client
URL = "https://mcp.gscwizard.com/mcp"
HEADERS = {"Authorization": "Bearer gscw_live_..."}
grok = OpenAI(api_key=os.environ["XAI_API_KEY"], base_url="https://api.x.ai/v1")
async with streamablehttp_client(URL, headers=HEADERS) as (read, write, _):
async with ClientSession(read, write) as session:
await session.initialize()
tools = [
{
"type": "function",
"function": {
"name": t.name,
"description": t.description,
"parameters": t.inputSchema,
},
}
for t in (await session.list_tools()).tools
]
completion = grok.chat.completions.create(
model="grok-4", # or whichever Grok model you use
messages=[{"role": "user", "content": "Which pages on example.com lost the most clicks in the last 28 days?"}],
tools=tools,
)
for call in completion.choices[0].message.tool_calls or []:
result = await session.call_tool(call.function.name, json.loads(call.function.arguments))
print(result.content)If you would rather not speak MCP at all, the same keys, scopes and rate limits back a plain REST surface at https://mcp.gscwizard.com/v1, with a generated OpenAPI 3.1 document at /v1/openapi.json you can hand to any model as a function-calling spec.
Check the URL first: it is https://mcp.gscwizard.com/mcp, with the /mcp path - the bare domain serves the documentation site, not the protocol. Custom connectors are available on every Grok plan, including the trial, and are managed at grok.com/connectors - so a missing connection is the endpoint or the sign-in, not your subscription. If the connector is added but no tools show up, remove it and add it again so the discovery step reruns, and make sure the sign-in actually completed.
Every analysis runs server-side against the data warehouse, so Grok reads a finished result instead of doing arithmetic over raw rows.
One Search Console connection, every assistant. Pick your client and follow the step-by-step guide.
The App Store app, a custom developer-mode connector, or the Responses API.
Read the guideAdd a custom connector in the Claude app, or one command in Claude Code.
Read the guideGemini CLI with a bearer header, Gemini Enterprise, or the Gemini API.
Read the guideThe endpoint, both auth modes, and copy-paste configs for every client.
Read the guideSee also the Google Search Console MCP overview, the GA4 MCP, and the ChatGPT app.
Full setup reference, including the Bot instructions and troubleshooting: the Grok documentation.
Yes, through a custom MCP connector. Go to grok.com/connectors, click New Connector and select Custom, then enter https://mcp.gscwizard.com/mcp. Grok discovers the tools the GSC Wizard MCP exposes and makes them available in conversations alongside the built-in and catalog connectors. Custom connectors are available on every Grok plan, including the trial.
No. The connector completes the authentication itself: the GSC Wizard MCP supports OAuth 2.1 with dynamic client registration, so Grok registers itself and sends you through a sign-in and consent screen. There is no client ID or secret to paste. API keys are for config-file clients such as Claude Code, Cursor, VS Code, Windsurf and Gemini CLI, which send a static Authorization: Bearer header instead.
Yes, Grok connects to the URL from its own infrastructure. That is only a problem for servers running on your laptop, which need a tunnelling service. GSC Wizard is hosted at https://mcp.gscwizard.com/mcp, so there is nothing to tunnel.
The consent screen lists exactly the scopes the client asked for, and you approve or deny them. For a connection that is read-only by construction, create a read-scoped API key under Account → API keys and use it in a client that accepts a bearer header: a read key physically cannot invoke a write tool. Every mutation made with a read-and-write grant is audit-logged either way.
Anything the MCP exposes: search analytics, top queries and pages, ranking changes, decay and cannibalization, CTR curves, opportunity scoring, bulk URL inspection, indexing tracking, IndexNow submissions, Bing Webmaster data, GA4 metrics, and full SEO reports. The analysis runs server-side, so Grok reads a finished result rather than doing arithmetic over raw rows.
No. Reads are served from a ClickHouse warehouse holding the full row set, so there are no export row caps and no GSC API paging. The warehouse lags roughly two days behind, and each response reports the date it is settled through.
A Grok Bot is a persistent agent with a fixed job and its own instructions, tools and routines. An SEO analyst Bot with the GSC Wizard connector attached keeps the analysis rules in place across every conversation, so you do not restate them. Bots can be shared by link, which copies the configuration only, never your credentials or conversation history.
Yes. Grok Build supports remote MCP servers, so you can register https://mcp.gscwizard.com/mcp and use Search Console data from the terminal while you work on a site. Skills and slash commands can wrap the common analyses so a whole workflow runs from one command.
Connect your properties, add one custom connector, and ask about your site in plain language.