GEO Basics · Product Update

Use Mettle directly from ChatGPT, Claude, Claude Code, and Codex with our MCP

A practical guide to using Mettle from ChatGPT, Claude, Claude Code CLI, and OpenAI Codex: how to enable the MCP connector in each environment, what to ask first, and how to turn brand health, competitor, and citation-gap data into weekly GEO actions.

May 15, 202611 min readBy the Mettle team

Reader guide

Turn this article into action

Plain-English point

A practical guide to using Mettle from ChatGPT, Claude, Claude Code CLI, and OpenAI Codex: how to enable the MCP connector in each environment, what to ask first, and how to turn brand health, competitor, and citation-gap data into weekly GEO actions.

Best use

Use this article when you are working on geo basics and need a concrete GEO next step.

Mettle MCPMettle ChatGPT connectorMettle Claude connectorMettle Claude Code MCPMettle Codex MCP

TL;DR: Mettle's MCP connector lets you bring real GEO data into ChatGPT and Claude. You can ask your assistant to check brand health, compare competitors, find citation gaps, and run prompt sets without copying screenshots or rebuilding context by hand. The real value is not "chat with a dashboard." It is turning AI search visibility into a weekly working session your team can actually use.

Why we built this

Most teams already use ChatGPT or Claude when they are thinking through strategy.

They paste a campaign idea. They ask for a content brief. They summarize a client call. They debate what to ship next.

But when the conversation turns to AI search visibility, the assistant usually has no real evidence. It does not know whether your brand is being mentioned. It does not know which competitors are winning. It does not know which domains are shaping the answer.

So the workflow becomes clumsy:

  1. Open a dashboard.
  2. Copy a few numbers.
  3. Paste them into the chat.
  4. Ask for interpretation.
  5. Realize one more data point is missing.
  6. Go back to the dashboard.

That is the problem Mettle's MCP connector is designed to remove.

With MCP, ChatGPT and Claude can call approved Mettle tools directly. Your assistant can pull the latest brand health, competitor intel, and citation gaps while you stay inside the conversation.

Key takeaway

The best use of Mettle MCP is not asking, "What does this metric mean?" The better use is asking, "Given the current evidence, what should we do this week?"

The bigger change behind the feature

Search is becoming less like a list of links and more like an answer layer.

Buyers ask questions such as:

  • "What are the best tools for this problem?"
  • "Which providers are trusted in this market?"
  • "How does this company compare to its competitors?"
  • "What should I shortlist?"

Those questions used to lead people through search results. Increasingly, they lead through AI-generated answers.

Gartner has predicted that traditional search engine volume will drop 25% by 2026 as AI chatbots and virtual agents take over more discovery behavior. Whether the exact number lands higher or lower, the direction is clear: more decisions are being shaped before the buyer reaches your site.

That creates a practical question for every founder, marketer, and agency:

When AI systems explain your category, are you part of the answer?

Mettle measures that. MCP makes it easier to work with that measurement in the place where decisions already happen.

What MCP means for a Mettle user

MCP stands for Model Context Protocol. In plain English, it lets an AI assistant connect to external tools and data sources through a controlled interface.

For Mettle users, the simple version is:

You can use Mettle directly from ChatGPT or Claude.

That means your assistant can:

  • list the brands in your Mettle account
  • check mention rate, sentiment, share of voice, and total mentions
  • compare competitor visibility
  • find domains that cite competitors but not you
  • start an active prompt set run

This changes the role of the assistant. It is no longer guessing from generic marketing knowledge. It is working from your Mettle account.

AvoidAsk Claude to brainstorm why a competitor might be showing up more often.

PreferAsk Claude to use Mettle, pull competitor intel, find citation gaps, and turn the evidence into a weekly action plan.

Interactive mental model

What happens when you use Mettle from ChatGPT or Claude?

Click through the four-step loop. This is the simplest way to explain MCP to a teammate: the assistant connects, gathers evidence, helps decide, then measures the next run.

Step 1

Connect

Add Mettle as a remote MCP server in ChatGPT or Claude.

Use https://api.mettlegeo.com/mcp, complete the Mettle sign-in flow, and approve access for the assistant.

Human question

Is this the right Mettle account and brand?

AI instruction

Call list_brands first and wait for the user to choose the brand.

How to enable Mettle MCP

Mettle works as a remote MCP server across ChatGPT, Claude, Claude Code, and OpenAI Codex. Pick the environment you use:

ChatGPT and Claude (web)

Use this server URL:

https://api.mettlegeo.com/mcp

The setup flow:

  1. Open the connector or MCP settings in ChatGPT or Claude.
  2. Add a new remote MCP server.
  3. Enter https://api.mettlegeo.com/mcp.
  4. Start the connection flow.
  5. Sign in to Mettle when the authorization page opens.
  6. Approve access.
  7. Return to ChatGPT or Claude and verify the connection.

Claude Code (CLI)

If you use Claude Code in the terminal, register Mettle with a single command:

claude mcp add --transport http mettle-mcp https://api.mettlegeo.com/mcp

That registers the Mettle MCP server under the name mettle-mcp. From any Claude Code session you can now ask:

Use Mettle to list the brands in my account.

To confirm the server is registered:

claude mcp list

OpenAI Codex (CLI)

Codex picks up MCP servers from its configuration file. Add the following entry to your ~/.codex/config.toml (or the project-level .codex/config.toml):

[[mcpServers]]
name = "mettle-mcp"
transport = "http"
url = "https://api.mettlegeo.com/mcp"

Restart Codex and verify with:

Use Mettle to list the brands in my account.

Use this as your first test regardless of which environment you chose:

Use Mettle to list the brands in my account.

If the connection works, the assistant should return the brands available in your Mettle account. From there, use the exact brand name in follow-up prompts.

Tip:

If you manage multiple brands or clients, always start with "list my brands." It prevents avoidable matching errors and keeps the rest of the session cleaner.

How authentication works

Mettle uses an OAuth-based authorization flow for MCP.

That means ChatGPT or Claude does not need your Mettle password. The assistant starts the connection request, Mettle asks you to sign in, and then Mettle issues an access token for the approved MCP client.

The token lets the assistant call approved Mettle tools through the protected /mcp endpoint. If access expires or is revoked, the assistant can restart the authorization flow.

The practical benefit is simple: you do not need to paste exports, API keys, or screenshots into a chat just to get a useful answer.

The first five prompts to try

Once Mettle is connected, do not start with a vague prompt like "analyze our GEO."

Start with smaller questions that build the context step by step.

Prompt playbook

Four prompts that make the connector useful fast

These prompts are written to be explicit for both people and AI assistants: first gather evidence, then explain what to do with it.

Prompt 1

Start clean

Use this at the beginning of a session so the assistant does not guess which brand to analyze.

Use Mettle to list my brands. Then ask me which one to analyze.

Prompt 2

Weekly pulse

Use this when you want a short, decision-ready readout instead of raw metrics.

Use Mettle to check brand health and competitor intel for [brand]. Tell me what improved, what weakened, and what deserves attention this week.

Prompt 3

Citation action

Use this to turn citation gaps into outreach or content work.

Use Mettle to find citation gaps for [brand]. Group the domains into quick wins, strategic publications, and low-priority sources. Recommend one action for each top source.

Prompt 4

Run review

Use this before or after running prompts so the team knows what to inspect.

Use Mettle to run the active prompt set for [brand]. Give me the run ID and a checklist for what to inspect when results are ready.

1. Confirm the brand

Use Mettle to list my brands. Then ask me which one to analyze.

This is boring, but it saves time. It makes sure the assistant is using the right account object before it starts reasoning.

2. Check health before strategy

Use Mettle to check brand health for [brand]. Summarize the last 7 days and last 30 days. Focus on mention rate, sentiment, share of voice, and total mentions.

What to look for:

  • mention rate rising or falling
  • sentiment dropping even if mentions are stable
  • share of voice changing faster than total mentions
  • a mismatch between the 7-day and 30-day view

That mismatch is useful. If 30 days looks fine but 7 days looks weak, something recent may be moving against you.

3. Find the competitor pattern

Use Mettle to get competitor intel for [brand]. Which competitors appear most often, and where is our visibility weakest by comparison?

Do not just ask who is first. Ask why the pattern matters.

Good follow-up:

Based on this competitor pattern, which competitor should we study first and what source advantage might they have?

4. Turn citation gaps into outreach

Use Mettle to find citation gaps for [brand]. Group the domains into quick wins, strategic publications, and probably-not-worth-it sources.

This is where the work becomes concrete. A citation gap is not just a missing link. It is a clue that a source is helping competitors become more believable in AI answers.

Ask the assistant to separate the gap list into actions:

  • update an existing profile
  • pitch a comparison page
  • publish a data asset
  • earn a review
  • ignore low-quality domains

5. Run the next prompt set

Use Mettle to run the active prompt set for [brand]. Give me the run ID and a checklist for what to inspect when results are ready.

The checklist matters because running prompts is not the final step. The useful step is reviewing what changed: mentions, competitors, citations, and framing.

A weekly GEO operating workflow

Here is the workflow we recommend for teams that want to make this useful every week.

Monday: check the market pulse

Prompt:

Use Mettle to check brand health and competitor intel for [brand]. Give me a one-page Monday pulse: what improved, what weakened, and what deserves attention this week.

Decision to make:

Are we defending visibility, gaining visibility, or losing ground?

If mention rate is stable but competitor pressure is rising, do not relax. That usually means the market is getting more crowded around the answer, even if your own metric has not collapsed yet.

Tuesday: inspect citation gaps

Prompt:

Use Mettle to find citation gaps for [brand]. Prioritize the top 10 domains by likely impact. For each one, suggest the most realistic way to earn presence there.

Decision to make:

Which source would change the answer if we appeared there?

The highest-count domain is not always the best target. A niche industry source that appears across several commercial prompts may matter more than a broad directory that shows up once.

Wednesday: choose the content action

Prompt:

Based on the Mettle health, competitor, and citation-gap data, recommend one content asset we should create or improve this week. Give me the angle, target prompt cluster, outline, and proof points needed.

Decision to make:

What asset gives AI systems a clearer reason to mention us?

That might be a comparison page, FAQ section, case study, technical guide, local market page, or original benchmark.

Thursday: prepare the team message

Prompt:

Turn this Mettle analysis into a short internal update for the team. Include the risk, the opportunity, this week's action, and how we will measure whether it worked.

Decision to make:

Can the team understand the GEO work without needing the whole dashboard?

This is where MCP is especially useful for agencies and founders. It converts data into a message people can act on.

Friday: run or review prompt sets

Prompt:

Use Mettle to run the active prompt set for [brand], or tell me what to review if a recent run is already available.

Decision to make:

Did the action move the answer layer, or do we need another iteration?

A stronger strategy prompt

Use this when you want a serious diagnosis instead of a generic marketing plan:

You are our GEO strategist. Use Mettle as the source of truth before making recommendations.

Analyze [brand] in this order:
1. brand health
2. competitor visibility
3. citation gaps
4. recommended next prompt set or follow-up check

Give me:
- the business risk in plain English
- the strongest competitor signal
- the most important missing citation source
- the one action we should take this week
- the metric we should watch after the action

Do not give generic SEO advice unless the Mettle data points to it.

This prompt is useful because it forces the assistant to move from evidence to decision. It also stops the assistant from filling the page with obvious advice like "publish high-quality content."

What good answers should look like

When the connector is being used well, the assistant's answer should be specific.

Weak answer:

You should improve your online presence and create more authoritative content.

Useful answer:

Your 7-day mention rate is weaker than the 30-day view, while two competitors are appearing more often. The top citation gaps are domains that already cite competitors. This week, prioritize one comparison asset and one outreach target rather than publishing another general blog post.

The difference is evidence.

Key takeaway

MCP does not make a bad prompt strategic. It gives a good prompt something real to work with.

Common mistakes

Mistake 1: asking for a full strategy too early

If you ask for a full strategy before the assistant has pulled brand health, competitors, and citation gaps, you will get generic advice.

Better:

First use Mettle to gather the evidence. Then create the strategy.

Mistake 2: treating citation gaps as a backlink list

Citation gaps are not only SEO targets. They are answer-shaping sources.

Ask:

Which of these citation gaps is most likely to change how AI systems describe us?

Mistake 3: ignoring sentiment

Mention rate can improve while sentiment weakens. That is not a win. It may mean your brand is being included with caveats, limitations, or weaker framing.

Ask:

If our mention rate and sentiment are moving in different directions, what does that suggest?

Mistake 4: running prompt sets without a review plan

Running a prompt set is only useful if you know what you are looking for afterward.

Ask:

Before running this prompt set, tell me what we should inspect when the results come back.

Quick troubleshooting checklist

If the connector does not work on the first try, check these points:

  • Wrong URL: use https://api.mettlegeo.com/mcp, not the dashboard URL.
  • Not signed in: complete the Mettle login step during authorization.
  • No brands returned: make sure your Mettle account has at least one tracked brand.
  • Brand not found: ask Mettle to list brands first, then reuse the exact brand name.
  • Prompt set not found: check that the brand has an active prompt set before asking the assistant to run it.
  • Expired access: remove the connector and reconnect if ChatGPT or Claude reports an authorization error.

Where this fits in the Mettle platform

The MCP connector sits on top of the broader Mettle workflow.

Mettle still helps you monitor:

  • brand mentions across AI-generated answers
  • competitor presence in commercial prompt clusters
  • citation sources that influence answers
  • sentiment and answer framing
  • prompt-set performance over time
  • multilingual and regional visibility patterns
  • technical readiness signals such as crawl access, structured data, and llms.txt

The dashboard remains the place for deeper review. MCP is the layer that lets you bring those signals into planning, client conversations, and day-to-day decisions.

The weekly question

If you remember only one prompt, use this every week:

Use Mettle to analyze [brand]. Tell me whether we are becoming more or less visible in AI answers, which competitor is creating the most pressure, which citation gap matters most, and what we should do this week.

That is the real job.

Not more dashboards. Not more screenshots. A tighter loop between evidence and action.

Sources and references

Questions this article answers

What is Mettle MCP?

Mettle MCP is a custom Model Context Protocol connector that lets ChatGPT, Claude, Claude Code, and OpenAI Codex connect to a Mettle account and use approved Mettle tools for GEO analysis.

What can I do with Mettle from ChatGPT or Claude?

You can list tracked brands, check brand health, compare competitor visibility, find citation gaps, and run an active prompt set from your AI chat or CLI session.

How do I enable Mettle MCP in Claude Code?

Run: claude mcp add --transport http mettle-mcp https://api.mettlegeo.com/mcp. That registers Mettle for all Claude Code sessions. Verify with: claude mcp list.

How do I enable Mettle MCP in OpenAI Codex?

Add a [[mcpServers]] block to ~/.codex/config.toml with name = "mettle-mcp", transport = "http", and url = "https://api.mettlegeo.com/mcp". Restart Codex to apply.

Why connect Mettle to ChatGPT or Claude?

Connecting Mettle to your AI assistant lets teams discuss strategy, retrieve live GEO data, and turn AI search visibility signals into actions without manually copying metrics between tools.

See what AI engines say about your site

Run a Mettle audit to check whether ChatGPT, Gemini and Claude can read your pages, and where competitors get cited instead.

Run an audit