Salespeak MCP Server: Ask Your AI Assistant About Your AI Traffic

You've got AI agents crawling your website every day. ChatGPT, Perplexity, Claude, Gemini. They're reading your pages, extracting your content, and answering questions about your brand. But until now, getting visibility into that behavior meant logging into a dashboard, clicking through filters, and building reports manually.

The Salespeak MCP Server changes that. It connects your Salespeak analytics directly to any AI assistant that supports the Model Context Protocol, so you can ask questions like "Which AI agents visited my site this week and what did they focus on?" and get an actual answer. In your AI assistant. In natural language.

No dashboard. No export. Just a conversation.

What the Salespeak MCP server actually does

MCP (Model Context Protocol) is an open standard, originally created by Anthropic, now governed by the Agentic AI Foundation under the Linux Foundation. It lets AI assistants connect to external data sources. Think of it as a structured API that AI models can use natively. Over 97 million monthly SDK downloads as of late 2025, adopted by every major AI platform. It's not experimental anymore. (If you want the deeper technical context, we covered how MCP fits into the agentic web architecture and how NLWeb uses MCP endpoints in earlier posts.)

The Salespeak MCP Server exposes three categories of data to any connected AI assistant:

1. AI agent activity on your website: which agents visit, what they read, how often, and what they appear to be doing
2. Content discovery tracking: whether AI agents have found your new content and how they're interpreting it
3. AI chat conversation analytics: what visitors are asking your on-site AI assistant and what patterns are emerging

Each of these becomes queryable through natural language once you connect the server to Claude, ChatGPT, or any MCP-compatible client.

1. Track AI agent activity and intent

The most common question marketers ask once they realize AI agents are visiting their site: "Which ones, and what are they doing?"

With the MCP server connected, you can ask your AI assistant directly:

• "Break down AI agent traffic by agent type (ChatGPT, Perplexity, Claude, Gemini) and summarize their browsing behavior."
• "Which AI agents are visiting my product pages vs. blog content vs. documentation?"
• "Show me the trend in AI agent volume over the past 30 days."

The server returns structured data that your AI assistant can reason about. You'll see agent volume trends, top visited pages, content category breakdowns, and inferred browsing intent, whether agents appear to be doing general research, product comparisons, or pulling specific data points.

This matters because AI agent behavior isn't uniform. Perplexity tends to hit different pages than ChatGPT. Claude's browsing patterns look different from Gemini's. Understanding those differences tells you which content is getting surfaced in AI-generated answers and which content is being ignored. (We wrote about why this visibility matters in What ChatGPT Says About Your Product Is the New KPI.)

2. Monitor whether AI agents find your new content

You published a new product page last Tuesday. A pricing update on Thursday. A blog post on Friday. Did any AI agents pick them up?

That question used to be unanswerable. Now you can ask:

• "Have AI agents found my latest content updates from last week?"
• "Which new pages were discovered by AI agents in the last 7 days, and what key topics did they extract?"
• "Are AI agents misclassifying any of my newly published pages?"

The server tracks discovery events (when an AI agent first hits a new URL) and cross-references against the topics it appears to extract. This surfaces mismatches early. If your enterprise pricing page is being categorized as "SMB solutions" by AI agents, you want to know that before it shows up as a wrong answer in ChatGPT.

Common outputs include newly discovered URLs, extracted key topics, summarization accuracy signals, and misclassification alerts. The kind of data that would take hours to compile from raw logs, delivered in a sentence.

3. Analyze AI chat conversations

If you're running Salespeak's on-site AI chat assistant, the MCP server opens up your conversation data to analysis you can't easily do in a traditional dashboard.

• "What are visitors asking my AI chat assistant most often this month?"
• "Summarize the top questions and map them to buyer journey stages."
• "What topics are coming up in chat that we don't have content for?"

The server clusters conversations by theme, identifies recurring questions, and maps them to funnel stages (awareness, consideration, decision). It also flags conversion blockers: questions where visitors seem ready to buy but can't find the answer they need.

This is where the MCP server becomes a content strategy tool. Instead of guessing what your next blog post should cover, you can ask your AI assistant "What questions are visitors asking that our content doesn't answer?" and get a data-backed answer. It's the same principle behind Answer Engine Optimization (understanding what AI systems need from your content) but applied to your own site's chat data.

How to set it up

The Salespeak MCP Server uses Streamable HTTP transport and OAuth 2.1 authentication. Setup takes a few minutes depending on your client.

Connecting to Claude Desktop

Add the server to your Claude Desktop MCP configuration:

{
  "mcpServers": {
    "salespeak": {
      "url": "https://platform.salespeak.ai/mcp",
      "auth": {
        "type": "oauth2",
        "client_id": "YOUR_CLIENT_ID",
        "client_secret": "YOUR_CLIENT_SECRET"
      }
    }
  }
}

Your client ID and secret are available in your Salespeak dashboard under Settings, Integrations, MCP Server.

Connecting to ChatGPT

ChatGPT supports MCP connections through its desktop app. Add the Salespeak server URL and authenticate with your OAuth credentials. The process is similar: point it at https://platform.salespeak.ai/mcp and authorize access.

Other MCP clients

Any client that supports Streamable HTTP MCP transport will work. That includes Cursor, VS Code with Copilot, and custom agents built with the MCP SDK. The server endpoint and OAuth flow are the same regardless of client.

What you can't do (yet)

The MCP server is read-only. You can query analytics and pull reports, but you can't configure your AI chat assistant, modify agent rules, or change settings through MCP. Those actions still require the Salespeak dashboard.

The server also doesn't expose individual visitor-level data through MCP connections. It returns aggregated analytics (agent volumes, content trends, conversation themes) not individual session records. This is a deliberate security decision. Independent security scans have found that roughly a third of MCP servers lack authentication entirely. We're not exposing PII through a protocol that most teams are still learning to secure.

Why this matters for your GTM stack

The typical workflow before MCP: open Salespeak dashboard, filter by date range, export data, paste into a doc, share with the team. That's fine for weekly reporting. It's terrible for quick questions.

With the MCP server, your AI assistant becomes the interface. A CMO preparing for a board meeting can ask "What's our AI visibility trend over the past quarter?" mid-conversation. A content lead can check "Did AI agents pick up the blog post I published yesterday?" without switching tools. A RevOps manager can ask "Which AI agents are hitting our pricing page most?" and get an answer before their coffee gets cold.

The data was always there. MCP makes it accessible where you're already working, inside the AI assistant you use every day.

Get connected

The Salespeak MCP Server is available to all Salespeak customers. Find your credentials in Settings, Integrations, MCP Server, connect to your preferred AI assistant, and start asking questions.

If you're not a Salespeak customer yet and want to see what AI agent analytics look like when they're queryable through natural language, book a demo. We'll connect the MCP server live and show you what AI agents are already doing on your site.