Key takeaways
- Promptwatch's MCP server connects your AI assistant (Claude, Cursor, or any MCP-compatible client) directly to your GEO data, so you can run Answer Gap Analysis in natural language without switching tabs.
- The integration exposes tools for gap analysis, prompt visibility checks, citation lookups, and competitor comparisons -- all queryable through conversation.
- Setup takes about 10 minutes: install the MCP server, add your API key, configure your client, and start asking questions.
- The real value isn't the setup -- it's the workflow change. When gap analysis lives inside your writing environment, you actually use it.
- This guide covers setup, practical prompt patterns, and how to turn what you find into content that gets cited.
If you've used Promptwatch for any length of time, you've probably run into the same friction: you're writing an article in Claude or drafting a brief in Cursor, you want to check which prompts you're missing, and then you have to stop, open a new tab, log in, navigate to Answer Gap Analysis, run the query, copy the results, and paste them back into your writing environment.
It's not a huge deal. But it breaks flow. And anything that breaks flow means you do it less often.
Promptwatch's MCP integration fixes this. Once it's set up, you can ask your AI assistant "what prompts are my competitors ranking for that I'm not?" and get real data back, right inside the conversation. No tab switching. No copy-pasting. Just a question and an answer.
Here's how to set it up and actually use it.
What MCP is and why it matters here
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and data sources. Instead of an AI model working only with what's in its context window, MCP lets it call out to live APIs, databases, and services mid-conversation.
Think of it like giving your AI assistant a set of specialized tools it can reach for when needed. When you ask Claude "what content gaps do I have for the query 'best project management software'?", it can use the Promptwatch MCP tool to actually query that data rather than guessing.
The protocol was developed by Anthropic and has been adopted broadly. Claude Desktop supports it natively. Cursor has MCP support built in. A growing number of other clients (Windsurf, Continue, custom agent setups) support it too.
For GEO work specifically, MCP is a natural fit. The questions you ask during content planning -- "who's getting cited for this prompt?", "what topics am I missing?", "how does my visibility compare to competitor X?" -- are exactly the kind of questions that benefit from live data rather than a model's training knowledge.
What Promptwatch's MCP exposes
Before getting into setup, it's worth knowing what you're actually getting access to. The Promptwatch MCP server exposes several tools your AI assistant can call:
- Answer Gap Analysis: Given a topic or prompt, returns the queries your competitors appear in that you don't. This is the core use case.
- Prompt visibility lookup: Check whether your brand appears in AI responses for a specific prompt, and if so, in which models.
- Citation source data: See which pages, domains, and content types are being cited for a given query across ChatGPT, Perplexity, Claude, and others.
- Competitor comparison: Pull a side-by-side of your visibility vs a named competitor across a prompt set.
- Prompt volume and difficulty: Get volume estimates and difficulty scores for prompts so you can prioritize what to work on.
Not every tool is available on every plan. The Essential plan ($99/mo) covers the basics. Crawler logs, city-level tracking, and deeper competitor data come in at Professional ($249/mo) and above.
Setting up the MCP integration
Step 1: Get your API key
Log into Promptwatch, go to Settings, and find the API section. Generate a new API key and copy it somewhere safe. You'll need it in the next step.
Step 2: Install the MCP server
Promptwatch's MCP server is distributed as an npm package. You need Node.js installed (v18 or later works fine).
npm install -g @promptwatch/mcp-server
If you'd rather not install globally, you can run it with npx:
npx @promptwatch/mcp-server
Step 3: Configure your MCP client
The configuration step varies depending on which AI assistant you're using.
Claude Desktop
Open your Claude Desktop config file. On macOS it's at ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows it's at %APPDATA%\Claude\claude_desktop_config.json.
Add the Promptwatch server to the mcpServers object:
{
"mcpServers": {
"promptwatch": {
"command": "npx",
"args": ["@promptwatch/mcp-server"],
"env": {
"PROMPTWATCH_API_KEY": "your_api_key_here",
"PROMPTWATCH_SITE_ID": "your_site_id_here"
}
}
}
}
Save the file and restart Claude Desktop. You should see a small tools icon appear in the interface confirming the server is connected.
Cursor
In Cursor, go to Settings > MCP and add a new server. The configuration is similar:
{
"name": "promptwatch",
"command": "npx @promptwatch/mcp-server",
"env": {
"PROMPTWATCH_API_KEY": "your_api_key_here",
"PROMPTWATCH_SITE_ID": "your_site_id_here"
}
}
Your site ID is in the Promptwatch dashboard under Settings > Sites. If you're tracking multiple sites, you can set up separate MCP configurations for each one, or pass the site ID dynamically in your prompts.
Other MCP-compatible clients
The pattern is the same: point the client at the @promptwatch/mcp-server command, pass your API key and site ID as environment variables. Any client that follows the MCP spec should work.
Step 4: Verify the connection
Once configured, ask your AI assistant something simple:
"Use Promptwatch to check my current visibility score."
If the connection is working, you'll see the assistant call the tool and return real data. If it fails, double-check that your API key is correct and that the npm package installed cleanly.
Running Answer Gap Analysis inside your AI assistant
This is where the setup pays off. Here are the prompt patterns that work well.
Basic gap analysis
"Using Promptwatch, run an answer gap analysis for the topic 'email marketing automation'. Show me the top 10 prompts where competitors are visible but I'm not."
The assistant will call the gap analysis tool, get the results, and present them in whatever format you ask for -- a table, a list, a prioritized brief. You can ask follow-up questions in the same conversation.
Prioritized gap analysis
"Run a gap analysis for 'project management software' and sort the results by prompt volume, highest first. Flag any prompts with a difficulty score under 40."
This is useful when you have a long list of gaps and need to decide where to start. High volume + low difficulty = the prompts worth targeting first.
Competitor-specific gaps
"Show me the prompts where [competitor name] appears in AI answers but I don't, specifically in ChatGPT and Perplexity."
You can filter by model, which matters because visibility isn't uniform across AI engines. A competitor might dominate Perplexity but be weak in Claude. Knowing that shapes where you focus.
Citation source lookup
"For the prompt 'best CRM for small business', what sources is ChatGPT currently citing? What type of content are they -- blog posts, comparison pages, Reddit threads?"
This tells you not just that you're missing, but what kind of content you need to create to fill the gap. If ChatGPT is citing Reddit threads and comparison pages, writing a generic blog post probably won't move the needle.
Gap-to-brief pipeline
This is the workflow that saves the most time. Once you have your gap analysis results, you can immediately ask the assistant to turn them into content briefs:
"Based on those gap analysis results, create a content brief for the highest-priority prompt. Include the target angle, key questions to answer, competitor sources currently being cited, and a suggested structure."
Because the gap data and the brief are in the same conversation, the assistant has full context. The brief it produces is grounded in actual citation data, not generic SEO advice.
Practical workflow: from gap to published content
Here's how a real session might look, start to finish.
Morning: identify gaps
You open Claude Desktop and ask for a gap analysis on your core topic area. You get back 15 prompts where competitors are visible and you're not. You ask the assistant to sort them by volume and flag the easy wins.
Still morning: pick your target
You pick the top prompt -- say, "what's the best tool for tracking AI search visibility" -- and ask for the citation sources. You see that Perplexity is citing a specific comparison article from a competitor site, and ChatGPT is pulling from a Reddit thread.
Midday: create the brief
You ask the assistant to build a content brief targeting that prompt, informed by what's currently being cited. The brief comes back with a suggested structure, the specific questions the content needs to answer, and notes on the angle that's missing from existing coverage.
Afternoon: write and publish
You write the article (or use Promptwatch's built-in AI writing agent for a first draft), publish it, and tag it in Promptwatch for tracking.
Next week: check results
You ask the assistant to pull your visibility data for that prompt and compare it to last week. If the new content is getting picked up, you'll see it.
The whole loop -- find gap, understand why, create content, track results -- happens without leaving your working environment.
Tips for getting better results
Be specific about which AI models you care about. "Show me gaps in ChatGPT" and "show me gaps across all models" return different data. If your audience uses Perplexity heavily, filter for that.
Use the difficulty scores. It's tempting to go after the highest-volume prompts, but a prompt with 10,000 monthly queries and a difficulty score of 85 is a long-term project. A prompt with 2,000 queries and a difficulty score of 25 can move in weeks.
Ask for query fan-outs. Promptwatch's prompt intelligence includes fan-out data -- how one prompt branches into related sub-queries. Asking "what are the fan-out queries for this prompt?" gives you a cluster of related topics to cover, not just one article to write.
Check Reddit and YouTube citations. If the assistant's citation lookup shows that AI models are pulling heavily from Reddit or YouTube for a topic, that's a signal. Publishing on those platforms (or getting mentioned there) might be faster than trying to rank a new page.
Run gap analysis before you write, not after. The most common mistake is writing content first and then checking visibility. Doing it in reverse -- checking what's missing, then writing to fill it -- is the whole point of having this data.
What this doesn't replace
The MCP integration is a workflow tool, not a strategy replacement. A few things to keep in mind:
The data is only as good as your prompt set. If you're only tracking 20 prompts in Promptwatch, your gap analysis will only surface gaps within those 20. Expanding your tracked prompt set (especially on Professional or Business plans) gives you a more complete picture.
Gap analysis shows you where you're invisible -- it doesn't tell you why. Sometimes you're missing because you don't have the content. Sometimes you have the content but it's not structured in a way AI models can parse. Sometimes the topic is dominated by a handful of high-authority sources that are hard to displace. The data surfaces the gap; you still have to diagnose the cause.
And finally: publishing content is necessary but not sufficient. AI models update their citations over time, but it's not instant. After publishing, give it a few weeks before drawing conclusions from the tracking data.
Comparison: using Promptwatch with MCP vs. without
| Workflow step | Without MCP | With MCP |
|---|---|---|
| Run gap analysis | Open dashboard, navigate to gap analysis, run query, copy results | Ask assistant directly, results in conversation |
| Check citation sources | Separate lookup in dashboard | Follow-up question in same conversation |
| Build content brief | Manual, based on copied data | Assistant builds brief with full gap context |
| Prioritize by volume/difficulty | Export data, sort in spreadsheet | Ask assistant to sort and filter inline |
| Track results | Check dashboard separately | Query tracking data in conversation |
The underlying data is identical. The difference is friction. Less friction means you do it more often, which means you actually close the gaps instead of just knowing about them.
Getting started
If you're already a Promptwatch user, the MCP setup is the fastest way to make your gap analysis habit stick. The 10-minute configuration pays back immediately in workflow efficiency.
If you're not yet using Promptwatch, the MCP integration is one piece of a broader platform that covers tracking, gap analysis, content generation, crawler logs, and traffic attribution. There's a free trial available at promptwatch.com.
The goal isn't to have a fancier dashboard. It's to close the gap between knowing you're invisible in AI search and actually doing something about it. The MCP integration just removes one more excuse not to.