If you spend most of your day in Claude (Claude Desktop or Claude Code inside Cursor / VS Code / iTerm), forcing yourself back to a separate Arenza dashboard tab to check brand visibility is friction. The Model Context Protocol (MCP) lets Claude talk directly to mcp.arenza.ai and pull live data — share of voice, wrong claims, opportunities, prompts — right inside the chat surface you already use.
This walkthrough takes about 5 minutes end to end. If you already have an Arenza account and Claude Desktop installed, skip to "Configure Claude Desktop" below.
What you can ask Claude after the install
- "What's my brand's share of voice across ChatGPT, Claude, Gemini, Perplexity, Copilot, and Grok this week?"
- "Show me the top 5 wrong claims AI assistants made about my brand last week — verbatim quote, severity, source LLM."
- "Which competitor is mentioned most often in Perplexity for the keyword \"best CRM for agencies\"?"
- "Suggest 10 buyer-perspective prompts for our new product line that we should be tracking."
- "Generate a canonical-fact article body to fix the wrong-claim finding F-2487 in zh."
Prerequisites
- Claude Desktop (claude.ai/download — macOS or Windows) OR Claude Code (https://claude.com/product/claude-code).
- An Arenza enterprise account (contact hello@arenza.ai — onboarding includes MCP server access).
- An Arenza account — on the first call Claude opens a browser to sign in and approve access, so there's no token to copy.
Configure Claude Desktop
Claude Desktop reads its MCP server list from a JSON config file. Open (or create) the file at:
- macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
- Windows: %APPDATA%\Claude\claude_desktop_config.json
Add the Arenza MCP server entry under "mcpServers":
{
"mcpServers": {
"arenza": {
"url": "https://mcp.arenza.ai/mcp"
}
}
}That's the whole config — just the URL, no token. Save the file. Quit Claude Desktop completely (Cmd+Q on macOS, not just close window) and reopen — the new MCP server only loads on cold start. The first time Claude calls an Arenza tool, a browser window opens to the Arenza sign-in and approve screen; approve once and the connection is remembered.
When Claude Desktop reopens, the bottom of the chat input shows a small "tools" icon. Click it: you should see "arenza" in the dropdown with its tools — list_brands, get_brand_overview, list_prompts, get_brand_corrections, get_brand_verified_info, verify_brand_claim, get_brand_discoverability, score_geo_article, and audit_site.
Configure Claude Code
Claude Code uses a workspace-level .mcp.json file (or your User Settings MCP block). The simplest install:
claude mcp add --transport http arenza https://mcp.arenza.ai/mcpOr, drop a .mcp.json into your project root for team sharing:
{
"mcpServers": {
"arenza": {
"url": "https://mcp.arenza.ai/mcp"
}
}
}No token in the file — nothing to keep out of git. Restart Claude Code and run /mcp in any chat. The first Arenza call opens your browser to approve access; after that "arenza" shows as connected.
Verify with a real prompt
In a fresh chat, paste:
Use the arenza tools to list my brands, then for the first brand call get_brand_overview and tell me the share of voice across every AI platform we track in plain English.
Claude will call list_brands → get_brand_overview, parse the response, and return something like: "Your brand 'Acme Co.' has share of voice 47% in ChatGPT, 22% in Gemini, and 51% in Perplexity. The lowest is Gemini — you're invisible in 78% of Gemini answers about your category."
If you instead see "I cannot connect to the arenza tools" — restart the Claude app and try the prompt again so the browser approve window can open. If it never appears, confirm the URL is exactly https://mcp.arenza.ai/mcp (the /mcp path, not /rpc) and that your browser isn't blocking the pop-up.
Troubleshooting
- "Tool not found" → the JSON config has a typo. Validate at https://jsonlint.com.
- "Unauthorized" → your sign-in session expired. Re-run the prompt and approve again in the browser when it opens.
- "Rate limited" → Free tier is 100 MCP calls/hour. Pro tier is 1,000/hour. See https://arenza.ai/llms.txt for the full ladder.
- Tool result is empty → the brand has not been scanned yet. From the dashboard, click "Scan now" or wait for the next weekly scheduled scan.
- Claude Desktop does not show the "arenza" server → you edited the wrong config file. The right path on macOS includes the literal "Application Support" directory name; not "ApplicationSupport".
How the connection is authorized
The setup above uses OAuth 2.0 — the URL is all you paste, and Claude handles the rest: Dynamic Client Registration (DCR) and PKCE mean there's no client_id to provision and no API key to manage. The browser approve screen is the only manual step, and it happens once. Full spec at https://mcp.arenza.ai/.well-known/oauth-authorization-server.
For agencies, each end-user approves their own access, so a junior only ever sees the brands they manage — there's no shared token to leak.
Headless agents, CI, and scripts
When there's no browser to run the approve step — a cron job, a CI pipeline, a server-side agent — use a personal access token instead. Create one at /settings/integrations and send it to the /rpc transport as Authorization: Bearer arn_pat_.... Same tools, no browser handshake.
Companion repo + screenshots
Working config files, the 5 example prompts as separate .md files, and screenshot guidance live in the open-source repo: https://github.com/arenza-ai/arenza-claude-tutorial — clone it, copy the example config, drop in your token, you are running.
Related guides
- Cursor + Arenza integration: https://arenza.ai/guides/cursor-mcp-arenza-quickstart
- n8n + Arenza for weekly automation: https://arenza.ai/guides/n8n-geo-automation-weekly-digest
- Slack alerts when AI misquotes your brand: https://arenza.ai/guides/slack-bot-ai-brand-misquote-alerts
- Why MCP-native matters for AI visibility tools: https://arenza.ai/guides/mcp-native-ai-visibility-architecture