Copy-pasting screenshots into a chat window is a slow way to work with GEO data. The Model Context Protocol (MCP) removes that step: your AI agent connects straight to a live data source and pulls numbers on demand. This guide shows you how to wire GEOly's GEO MCP server into Claude, Cursor, Codex, and Claude Code so you can ask "How visible is my brand for 'enterprise CRM' right now?" and get a real answer inside your editor, with no dashboard tab-switching.
By the end you will have MCP connected through OAuth or a read-only token, know which of the tools to call, and understand how the same toolset reaches you through the GEOly CLI for terminal and CI work.
Key takeaways
- MCP is an open standard that lets AI clients (Claude, Cursor, Codex, Claude Code) call external tools directly, so your agent reads live GEO data instead of stale exports. - GEOly ships a GEO MCP server at `app.geoly.ai/api/mcp` with up to 62 tools across two data planes: your own brand (KPIs, prompts, citations, competitors, audit, GA4) and industry-public data (brand rankings, category white space, Share of Card, citation sources). - You can connect with one line of config using either OAuth browser authorization or a read-only API token. - The same tools are available in the terminal through the GEOly CLI, so agents in Claude Code, Codex, n8n, or CI pipelines get identical numbers with zero drift. - MCP tools unlock from the Grow plan; every new account gets a free 3-day trial at app.geoly.ai.
Step 1: Understand what MCP gives you for GEO
MCP is often described as the USB-C port for AI apps. It is an open standard that lets a model talk to an external server through a defined set of tools, safely and predictably. Instead of you fetching a report and pasting it in, the agent calls a tool, receives structured JSON, and reasons over it.
For GEO work that means three practical wins. Your agent reads live visibility numbers rather than a week-old CSV. It stays context-aware, so while it edits a landing page it can check your current AIGVR for the target query. And it can run unattended, auditing pages or pulling competitor movement on a schedule.
GEOly exposes two data planes through one server. The own-brand plane covers your KPIs, prompt-level visibility, citations, competitors, the GEO Audit, and GA4 attribution, plus write actions like creating a prompt, topic, or competitor and triggering a monitoring run. The industry-public plane lets any agent query category rankings, white space, brand momentum, Share of Card, AI search demand, brand comparisons, and citation sources without configuring anything first.
Gotcha: MCP tool access starts on the Grow plan. On the free trial you can connect and explore, but confirm your plan covers the tools you plan to script against.
Step 2: Get your GEOly credentials
Sign in at app.geoly.ai (start the free 3-day trial if you are new). GEOly's MCP server supports two auth modes, and which you pick depends on the client.
For interactive desktop clients like Claude Desktop and Cursor, OAuth browser authorization is cleanest: the client opens a browser, you approve access, and no secret is stored in a config file. For headless or terminal use, generate a read-only API token in your account settings and pass it as a header or environment variable.
Gotcha: treat the read-only token like a password. Keep it out of committed config and rotate it if it leaks. Read-only means the agent cannot write, which is exactly what you want in CI.
Step 3: Connect the server in Claude and Cursor
Claude Desktop remains the reference client for local MCP use. Add the GEOly server to your MCP configuration pointing at `app.geoly.ai/api/mcp`, then restart the app so it loads the connection. On first call it will trigger the OAuth flow; approve it in the browser and the tools appear.
Cursor follows the same pattern from its MCP settings: add a server entry with the same endpoint and either OAuth or your token. Once connected, Cursor's agent can call GEOly tools while it writes code, so a prompt like "check my Share of Model for our top three competitors and adjust the comparison table on this page" runs end to end.
Codex and Claude Code connect the same remote endpoint through their own MCP configuration. Because every client hits the identical server, the tool names and metric definitions are the same everywhere.
Gotcha: if tools do not show up, the usual causes are a typo in the endpoint, an app that was not fully restarted, or a token pasted with a trailing space. Re-open the client and confirm the server shows as connected before debugging further.
Step 4: Call the right tool for the question
With the server live, phrase requests in plain language and let the agent map them to tools. Ask for your AIGVR and Share of Model on a set of prompts and it will pull your own-brand KPIs. Ask which domains ChatGPT cites for your category and it will hit the citation-source tools on the industry plane. Ask it to add a new monitored prompt and it uses a write action.
A good working pattern: start every session by having the agent discover your org and brand IDs, then scope subsequent calls to them. This avoids ambiguous results when your account holds more than one brand.
For a deeper tour of the server, its tool catalog, and worked agent prompts, read the companion piece on the [GEOly GEO MCP server for Claude Code, Cursor, and Codex](/blog/geo-mcp-server-ai-visibility-claude-code-cursor-codex).
Step 5: Move the same tools into the terminal with the GEOly CLI
The GEOly CLI is the remote MCP server projected into your shell. It is a single zero-dependency binary you install with one curl or PowerShell line, it prints clean JSON to stdout, and it returns branch-able exit codes. That makes it the right fit for Claude Code, Codex, n8n, and CI pipelines where you want scripting rather than a chat UI.
Because the CLI and MCP share one tool source on the server, adding a tool server-side needs no CLI release and the metric definitions never drift between the two. Download it and read the docs at www.geoly.ai/open/cli.
Gotcha: pipe CLI output to `jq` and check exit codes in scripts so a failed call fails the pipeline loudly instead of silently returning empty data.
How to verify it works
Run one read call and one scoped call. First, ask the agent for your account's brands; a correct setup returns your brand list. Then ask for AIGVR on a single known prompt and confirm the number matches what your GEOly dashboard shows for the same window. If both match, your connection, auth, and metric scoping are correct.
FAQ
How many tools does the GEOly MCP server expose? Up to 62, split between your own-brand data and the industry-public data layer.
Do I need to pay to use MCP? MCP tools unlock from the Grow plan. A free 3-day trial lets you connect and explore before committing.
Which clients are supported? Any MCP-capable client, including Claude Desktop, Cursor, Codex, and Claude Code, plus terminal and CI workflows through the GEOly CLI.
OAuth or token, which should I use? OAuth for interactive desktop clients so no secret is stored; a read-only token for headless, terminal, and CI use.
What is GEOly, exactly? A GEO data platform built around industry intelligence: it tracks any brand and whole categories across ChatGPT, Gemini, and Google AI, and exposes that data to your agents. See [what GEOly AI is](/blog/what-is-geoly-ai) or the guides from [GEOly AI](/blog/author/geoly-ai).

