gsc-mcp-connector
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@gsc-mcp-connectorshow my top 10 queries by clicks last month"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
gsc-mcp-connector
Self-hosted Google Search Console MCP server, deployable to Cloudflare Workers in 15 minutes. Plug it into ChatGPT, Claude, or any MCP-capable client and query your GSC data in natural language.
What this gives you
Once deployed, you get a private MCP endpoint that exposes 4 tools to your AI assistant:
list_sites— discover every property accessible to the authenticated userquery_search_analytics— clicks, impressions, CTR, position, filterable by query / page / country / device / search appearance / dateinspect_url— full URL Inspection API output (indexing status, canonical, mobile, AMP)list_sitemaps— every submitted sitemap and its processing status
You ask: "Quelles sont mes 50 requêtes avec la plus grosse perte de clics entre les 28 derniers jours et les 28 jours précédents ?" and the assistant pulls the data, computes the delta, and writes the analysis. No more SQL exports.
Related MCP server: gsc-mcp
How it works
ChatGPT/Claude ──OAuth──▶ Your Worker ──OAuth──▶ Google
│
└─ holds your Google refresh_token
(encrypted, in OAuth grant props)Two OAuth chains :
MCP client → Worker : ChatGPT/Claude do OAuth 2.1 + PKCE against your Worker. The login UI asks for a static "access key" you set at deploy time (the connector gate).
Worker → Google : after the access key check, the user is redirected to Google's consent screen to grant
webmasters.readonly. The resulting refresh token is stored in the OAuth grant; on every tool call, the Worker refreshes a fresh access token and calls GSC.
The user's Google account drives access — no Service Account, no GSC user-management dance, no permission propagation delays.
Prerequisites
Item | Cost | Required ? |
ChatGPT Plus / Pro / Team or Claude.ai Pro / Team | $20/mo+ | Custom MCP connectors are gated on paid plans. |
Cloudflare account | Free tier is enough | Yes |
Google Cloud project | Free | Yes — to create an OAuth Client |
Verified Search Console property | Free | Yes (you already have it) |
Node.js 20+ + | Free | Recommended for the secret-setting steps |
The Cloudflare Workers free tier (100k requests/day) is largely enough for personal SEO usage. No paid Cloudflare plan needed.
Quick start (~15 min)
1. Deploy the Worker
Click the Deploy to Cloudflare button at the top of this README. Cloudflare clones the repo into your account, installs deps, and gives you a public URL like https://gsc-mcp-connector.<your-subdomain>.workers.dev.
Note : at this stage, you'll be asked for
MCP_BEARER_TOKEN,GOOGLE_OAUTH_CLIENT_ID, andGOOGLE_OAUTH_CLIENT_SECRET. You don't have the Google ones yet — fillMCP_BEARER_TOKENwith any random hex string for now (you can change later), and paste anything in the two Google fields. We'll set them properly in step 4.If you'd rather skip the button, clone the repo locally, run
npm install, thennpx wrangler deploy.
After deployment, note your Worker URL. You will need it both in step 3 and in step 5.
2. Generate a connector access key
This is a static random string used as a gate before Google OAuth. Anyone using the connector must paste this string in the login UI.
openssl rand -hex 32Save the output — you'll set it as a secret and use it in ChatGPT/Claude.
3. Set up Google Cloud (OAuth Client)
Follow the step-by-step in docs/SETUP_GCP.md. Use your Worker URL from step 1 in the Authorized redirect URI. You'll end up with a Client ID and a Client Secret.
This is the longest step (~10 min the first time) but you only do it once.
4. Configure the Worker secrets
npx wrangler secret put MCP_BEARER_TOKEN
# paste the value from step 2
npx wrangler secret put GOOGLE_OAUTH_CLIENT_ID
# paste the Client ID from step 3
npx wrangler secret put GOOGLE_OAUTH_CLIENT_SECRET
# paste the Client Secret from step 3Or via the Cloudflare dashboard : Workers & Pages → your worker → Settings → Variables and Secrets → add each as type Secret.
Tip — encoding pitfall on Windows : if you pipe a file's content (e.g.
Get-Content | wrangler secret put) and that file has a UTF-8 BOM, the BOM ends up in your secret and breaks JSON parsing. The Cloudflare dashboard or interactivewrangler secret put(paste at the prompt) avoids this entirely.
5. Plug into ChatGPT (Plus/Pro/Team)
ChatGPT → Settings → Connectors → Add custom connector :
Name :
gscMCP Server URL :
https://YOUR-WORKER-URL/mcp(must end with/mcp)Authentication :
OAuthCheck "I understand and want to continue"
Create
A popup opens to your Worker's login UI. Paste your MCP_BEARER_TOKEN → click Continue with Google → → Google asks you to log in (use the account that owns your GSC property) and grant webmasters.readonly → you're redirected back to ChatGPT, connector is active.
In a new chat, enable the gsc connector in the toolbar, then ask: "List my Google Search Console sites". You should see your properties.
6. (Optional) Plug into Claude.ai (Pro/Team)
Settings → Integrations → Add custom integration → same URL, same flow.
Local development
git clone https://github.com/JuJu78/gsc-mcp-connector
cd gsc-mcp-connector
npm install
cp .dev.vars.example .dev.vars
# edit .dev.vars with your real Client ID + Client Secret + bearer token
npx wrangler devThe dev server runs on http://localhost:8787. Note that local dev cannot fully complete the Google OAuth flow because Google's redirect URIs require HTTPS. For a true end-to-end test, deploy to Cloudflare and test against the workers.dev URL.
Limitations
Read-only. Adding write operations (submit sitemap, request indexing) is left out by design — they're risky in an LLM context. Open a PR if you need them.
Single-tenant by design. One operator deploys, one bearer token gates the connector, the access is bound to whoever completes the Google OAuth dance. Multi-user SaaS-style is out of scope.
OAuth consent in Testing mode caps you at 100 test users (more than enough for personal/team use). For broader distribution you'd need to submit the app for Google verification (
webmasters.readonlyis a "sensitive" scope, requires manual review).GSC API quotas — 1200 queries/min/project, 30k/day. Plenty for interactive use.
Date range — GSC returns the last 16 months. Earlier dates error.
Troubleshooting
Symptom | Cause | Fix |
| Authorized redirect URI in GCP doesn't exactly match what the Worker sends | Verify |
| Logged-in account isn't in Test users of the OAuth consent screen | Add the Gmail in OAuth consent screen → Audience → Test users |
| Stale OAuth grant from a previous attempt | Delete the connector in ChatGPT and recreate it |
| The grant was created before you deployed v0.4+ | Delete the connector in ChatGPT/Claude and recreate it |
|
| Re-set the secret via interactive |
| Authenticated Google account has no GSC properties (or wrong account) | Verify which account you used in the Google consent step — it must own GSC properties |
ChatGPT says "no tools available" | URL doesn't end with | Server URL must be |
Stack
Cloudflare Workers + Durable Objects (via
agentsSDK ≥0.12)@cloudflare/workers-oauth-providerfor OAuth 2.1 + DCR + PKCE@modelcontextprotocol/sdkfor tool definitionsKV namespace
OAUTH_KVfor OAuth stateGoogle OAuth 2.0 flow signed natively via Web Crypto API (no Node deps)
Credits
Built by Julien Gourdon — SEO consultant exploring the intersection of search and AI.
License
MIT — see LICENSE.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/JuJu78/gsc-mcp-connector'
If you have feedback or need assistance with the MCP directory API, please join our Discord server